Honours and Masters project

 Title: Health and Social Challenges of Refugee Populations in Australia: A Data-Driven Investigation.

Keywords: Refugees, health outcomes, social challenges, data integration, policy analysis

Project Description:
Perform a comprehensive data-driven study on health and social challenges faced by refugee populations in Australia. The project integrates multiple datasets, applies advanced statistical analysis, and uses machine learning to detect patterns that inform policy and support services.

Objective:
To analyze and understand health and social factors affecting refugees to improve policy responses and support mechanisms.

Methods:

• Multi-source data integration
• Advanced statistical and exploratory analysis
• Machine learning pattern recognition

This project uses machine learning and predictive analytics to group customers based on their shopping habits using publicly available or synthetic transactional datasets. Students will clean and analyse purchase data, apply clustering algorithms such as K-Means and Hierarchical Clustering, and identify common product purchase patterns using association rule mining. The project aims to show how data-driven methods can help businesses better understand customer behaviour and design targeted marketing strategies.

 Title: Grouping Customers by Shopping Habits with Machine Learning

Description:
This project focuses on analysing customer purchase data to identify distinct groups of customers with similar shopping habits. Students will work with public or synthetic transactional datasets to clean and prepare the data, then apply clustering algorithms such as K-Means and Hierarchical Clustering to discover meaningful customer segments. The project also includes using association rule mining to find patterns in product purchases within each group. The results will help businesses create targeted marketing strategies based on customer behaviour.

Specialised project:

This project applies machine learning and predictive analytics to detect early signs of heart disease using publicly available cardiovascular datasets. Students will clean and analyse health data, apply algorithms such as Decision Trees and Random Forest, and identify key risk factors for heart disease. The project aims to show how data-driven methods can support early intervention and improve patient outcomes.

Title: Early Detection of Heart Disease Using Machine Learning and Predictive Analytics

Description:
This project focuses on developing machine learning models to detect early signs of heart disease from patient health data. Students will work with publicly available cardiovascular datasets to clean and preprocess data, select important features, and build predictive models using algorithms such as Logistic Regression, Decision Trees, and Random Forest. The aim is to identify patients at risk of heart disease early, enabling timely interventions and better health outcomes.

This research aims to bridge a critical accessibility gap in digital navigation tools by developing an inclusive, intelligent system that combines map services, street-level imagery, and large language models (LLMs). Current systems often fail to support marginalised users—such as older adults, people with vision impairments, or those with limited mobility—by overlooking nuanced environmental cues such as footpath obstructions, ramp availability, or visibility of building entrances. By democratising navigation, the project addresses both a technological and equity gap in urban mobility.

Urbanisation and climate change are accelerating environmental degradation, making cities critical battlegrounds for sustainability. While vast amounts of environmental data (e.g., CO₂ emissions, energy use, air quality, weather, etc.) are collected, extracting actionable insights remains a challenge due to data complexity, real-time processing demands, extensive human/expert involvement, and the need for predictive analytics.  This project aims to develop AI-powered Large Language Model (LLM) agents that autonomously interpret urban environmental data, detect anomalies, forecast trends, and provide decision-support for sustainable urban management.

In human-AI collaboration, it is essential for AI systems to understand and anticipate human behavior in order to coordinate effectively. Conversely, humans also form inferences about the agent’s beliefs and goals to facilitate smoother collaboration. As a result, AI agents should adapt their behavior to align with human reasoning patterns, making their actions more interpretable and predictable. This principle forms the foundation of transparent planning (MacNally et al, 2018).

A key prerequisite for transparent planning is a robust model of how humans interpret and infer goals from observed behaviors (i.e., a human model of goal recognition). Prior work has explored human goal recognition in controlled environments (Zhang et al, 2024) via manipulating factors such as timing and solvability in puzzle-like tasks. However, that work only focuses on passive settings. As an extension of goal recognition tasks, active goal recognition (AGR) offers a more realistic framework where observers are not passive but can interact with the environment to gather information actively. Therefore, there is a need to extend these models to account for human behavior in active goal recognition scenarios. Such an extension is critical for building AI systems that can both interpret and be interpreted by humans in interactive, real-world environments.

Since the COVID-19 pandemic there has been an increasing shift within higher education away from traditional lectures and towards asynchronous content delivery through pre-recorded videos. This has a number of benefits: students can consume content at their own pace, videos can be reused, and production value can be increased. However, academics typically have no training or experience in video production, so pre-recorded videos are most often just a simulacrum of a standard lecture (i.e., a slideshow with voiceover). Analytics and student feedback indicate that this style is not engaging.

Parallel to this shift has been the increasing popularity of educational “youtubers”. Many youtubers (e.g., CGP Grey, Vsauce, Kurzgesagt) are successful enough to earn a full time income, often for multiple people, from creating and distributing educational videos online.

This project aims to improve student engagement with asynchronous video content by identifying and emulating successful strategies used by educational youtubers. Possible activities include:

• a structured review of educational youtube content
• developing and evaluating educational videos
• building guidelines or a framework to help other academics improve their asynchronous content

Your contributions have the potential to have real impact. Any video resources developed will focus on introductory programming, and be distributed through the open education resource The Programmer's Field Guide (https://programmers.guide/), which is used by thousands of students each year. We also hope to engage with you as a person, and to help you develop in the areas you are wanting to focus on.

Access to education is an important issue. A major factor preventing access can be the cost of textbooks, which is a significant barrier for some students. Open Education Resources (OERs) are a popular option for reducing this financial burden, as they are free to any person with an internet connection. 

The Programmer’s Field Guide is an OER for learning introductory programming which has already been used by thousands of students. Written by experienced university educators, it is a resource for learning programming from the bottom-up, focusing on the concepts underpinning programming rather than the syntax of a specific language. 

In this project we aim to improve accessibility of The Programmer’s Field Guide (https://programmers.guide/). Possible activities include:

• improve the accessibility of the digital edition based on best practice recommendations for readers with vision or related impairments
• develop an automated way of converting the digital edition to a print on demand edition for readers who need or prefer physical materials

Our aim with this project is to allow a wider variety of readers to use the resource, which will continue to be updated and distributed freely. We also hope to engage with you as a person, and to help you develop in the areas you are wanting to focus on.

Introductory programming remains a significant challenge for many students. A large factor impacting success is each student's motivation to engage with assessment and practice exercises. One strategy for improving student engagement is to offer multiple assessment options. These should cover the same concepts, and be of equivalent difficulty, but be themed according to a variety of interests (e.g., games, data science, hardware). By providing such flexibility, students are empowered to engage in learning activities and assessment tasks which are more aligned with their interests and future aspirations.

This project aims to build and evaluate a framework for designing equivalent assessment options. Possible activities include:

• identifying appropriate measures of task complexity/difficulty
• building software tools to automate estimation of task complexity/difficulty
• developing guidelines for designing equivalent assessment options for introductory programming
• building equivalent assessment tasks with different theming for introductory programming
• evaluating task difficulty by comparing estimates with complexity of real student work

Your contributions have the potential to have real impact. Our goal is to integrate this work into FIT1045 Introduction to Programming, which is completed by thousands of students each year. We also hope to engage with you as a person, and to help you develop in the areas you are wanting to focus on.

Leveraging the FLoRA adaptive learning platform, we will conduct a five-phase research program combining experimental studies and advanced trace data analysis. Through time-stamped interaction data, we aim to detect behavioural signals of metacognitive disengagement using machine learning and time-series modeling techniques. These insights will inform the development of adaptive scaffolding tools that encourage students to monitor, evaluate, and adjust their learning strategies when using GenAI.

Australia’s cybersecurity infrastructure, particularly in IoT networks, must be strengthened to meet evolving standards set by international bodies like NIST and the NSA. This project will support Australian organizations in adapting to quantum-safe standards, ensuring the protection of sensitive data and critical system

With the rise of quantum computing, traditional cybersecurity mechanisms like RSA and ECC, commonly used to protect data, may soon become vulnerable. This project focuses on preparing IoT networks (such as critical infrastructure monitoring, fleet management, healthcare, and EV charging) for this shift by transitioning to post-quantum cryptographic (PQC) methods, which are resistant to quantum attacks. Key goals include developing IoT Networks in the above mentioned contexts, find out about all quantum-vulnerable security/mechanisms that might have been used in such networks, and perform a risk assessment for each application.

What will students learn in this project

Students are expected to learn about:

• How to setup/configure a working IoT network (both software and hardware) in different contexts and applications
• How to find/retrieve different security/privacy mechanisms/protocols used in a given IoT network.
• Perform risk assessment on all security/privacy mechanisms on different layers of IoT Networks, including hardware, software, and network traffics.

Skills Required

• Basic understanding of network protocols.
• Ability to work with constraint devices, routers, switches, and setup an IoT network.

Application Instructions

Please attach a copy of your CV and current transcripts.

Manual medical report writing is time-consuming and subject to variability. Recent advances in large language models (LLMs) create new opportunities for automating this process. This project explores using LLMs to generate medical reports from a very large dataset, aiming to streamline workflows and support clinical decision-making. Students will work on data preprocessing, model fine-tuning, and performance evaluation, contributing to advances in medical AI.

Using artificial intelligence software and unique algorithms for predictive analytics that incorporate modelling, machine learning, and data mining, we analyse, model, and build an individual’s baseline health profile against thousands (eventually millions) of similar people and their data points, along with decades of evidence-based medical and population research. Our previous work focused on the prediction of Diabetes Type Two – a major debilitating chronic disease, and a significant contributor to global deaths. This work then led to an understanding of the extended co-morbidities that both link and surround all chronic diseases and conditions, and how we might provide tools to allow individuals to understand their risk at any time in their lives. We aim to deliver integrated healthcare strategies that combine personalised guidance with ongoing maintenance, fostering a lifelong, adaptive personal health ecosystem.

As games have increased in complexity, so have the worlds in which players find themselves. The automap has become a staple feature of many video games; in some, the players is allowed to edit the map itself during play (e.g., to mark important locations or lay out a course). We know about what games offer players in terms of game maps, but not how players *actually use* those features. This project will collect data on players' cartography activities in games. 

The project will develop students' experiences with game design, visual analysis, questionnaire development, and reflexive thematic analysis. 

Note that Supervisor Toups Dugas does not employ any form of generative AI in her research; supervised students should expect that such tools will *not* be allowed for work on this project. 

Gender euphoria addresses times when one's lived experience aligns with their gender identity. This may be personal experiences of one's body, how one is treated by others, or through in-game experiences. Our prior research has developed an understanding of how gender euphoria comes through in video games from first-person research. This project will work toward developing and deploying a questionnaire to study this phenomenon in the wild and analyse the results. 

The project will develop students' experiences with game design, gender in design, questionnaire development, and reflexive thematic analysis. 

Note that Supervisor Toups Dugas does not employ any form of generative AI in her research; supervised students should expect that such tools will *not* be allowed for work on this project. 

What is a mixture model?

You may have learned about mixture models in a machine learning or statistics course. A mixture model with K component densities is defined by

• a set of K nonnegative mixture weights summing to one, and
• a corresponding set of K nonnegative component densities, each of which integrates to one.

The sum of the product of the mixture weights and component densities is guaranteed to be nonnegative and integrates to one, meaning it is a valid probability density.

What is a subtractive mixture model?

There are other ways to construct functions which are are nonnegative and sum to one (valid probability densities). One such way involves subtractive mixture models. It is possible to construct

• a set of K mixture weights, which are not necessarily nonnegative and do not necessarily sum to one, and
• a corresponding set of K nonnegative component densities, each of which integrates to one

such that the sum of the product of the mixture weights and component densities is nonnegative and integrates to one.

Sampling

Assuming we can sample from the component densities, it is easy to sample from a mixture model. First, sample a component index from a categorical distribution defined by the mixture weights. Then, sample from the component densities.

The same sampling technique does not apply to subtractive mixture models.

Pretrained models

The hidden layers of pretrained foundation models, such as ChatGPT, contain useful and abstract summaries of data. From an information-theoretic perspective, they might compress the data. From a machine learning perspective, they compute useful features of the data. From a statistics perspective, they might be sufficient statistics for a parameter of interest.

Probabilistic models

Among other things, probabilistic models can quantitatively describe the long-term frequencies of events (such as words or sentences) or beliefs about events. A classical approach for building such models is to apply the exponential function to a linear transformation of the statistic, and normalise the result.

This project involves building a system that processes IoMT data(such as heart rate, blood pressure, or glucose levels) from wearable devices to monitor patient health in real time.

The system uses machine learning to detect anomalies and alert healthcare providers or caregivers. It includes data preprocessing, model training, and a simple dashboard for visualization.

This project aims to develop a modular and distributed digital twin framework focused on simulating, monitoring, and analysing smart grid data. The framework will represent virtual models of grid components (e.g., loads, meters, nodes) and synchronise them with real-time or simulated data streams using distributed systems principles.

The testbed will support:

• Digital twin modeling of grid entities (smart meters).
• Distributed data ingestion and synchronization using messaging protocols.
• Time-series data storage for historical and real-time analysis.
• Interactive dashboards for monitoring grid behaviour.
• Basic analytics for load profiling and anomaly detection.

This foundational framework will be extensible for future integration with machine learning models, renewable energy systems, and real-world IoT data sources.

This is a research and development project focused on designing a Generative AI tool that supports equitable team practices in software development. The project combines qualitative research, such as persona profile creation, with AI prototyping to explore how GenAI can foster inclusion, improve team dynamics, and accommodate diverse working styles in technical environments. The outcome includes both a functional AI prototype and practical resources for inclusive collaboration.

This project focuses on implementing an AI-powered digital twin for intelligent electric vehicle (EV) traffic management in smart cities, utilising the YOLO algorithm. It develops a basic digital twin system designed to monitor and manage EV traffic in urban areas. The system detects and tracks EVs using feeds from traffic cameras. The digital twin simulates traffic flow, providing a visualisation of EV movement, congestion points, and route patterns.

This project develops a collaborative handover system using an augmented reality (AR) headset display to improve ICU nurse communication. It integrates in-situ AR visualisations and a cross-reality collaboration model to address cognitive challenges and data fragmentation in handovers. The prototype aims to enhance the accuracy and efficiency of handover for practical application in high-pressure settings. Broader impacts include applications in emergency care, rural health, and allied health professions, with potential for simulation training and multidisciplinary team support across diverse clinical environments.

This research addresses the challenges in the nurse handover process in critical care settings by designing an integrated solution that leverages immersive technologies. We propose an integrated approach consisting of the following innovative concepts: in-situ augmented reality (AR) overlays and a cross-reality collaborative model.

In-situ augmented reality overlays — The project will mainly explore the use of embedded visualisations (e.g., vital sign trends and digital anatomical models) in 3D space, displayed through augmented reality headsets at the bedside, to ensure data integrity over time and between nurses, assisting nurses in promptly identifying abnormal conditions, and enhancing clinical decision-making.

Cross-reality collaborative model — The project will also explore leveraging augmented reality to develop a cross-reality system that supports collaborative but remote and even asynchronous nurse handovers. Traditional handover requires nurses to be co-located, while AR could enable asynchronous information exchange (e.g., pre-recorded handover or a remote handover process for patient transfer) for shared understanding and flexible communication during the transition of care.

This project builds upon an existing safety inspection system framework, integrating augmented reality (AR), artificial intelligence (AI), and automated reporting. You will follow the framework to design, implement, and evaluate a fully functioning prototype system that supports safety inspections in real-world construction environments.

The proposed system leverages Apple Vision Pro to support AR-assisted inspection, allowing inspectors to track inspected areas using a virtual brush, capture live video streams and photos, and annotate findings through voice or text. These inputs are processed by an AI-assisted gap detection module, featuring image segmentation and gap classification models powered by Ultralytics YOLOv11. Finally, the inspection data is fed into a Python-based module that generates comprehensive inspection reports for safety analysis and documentation.

Through this project, you will work closely with real construction scenarios and contribute to the ongoing industry need for smarter, technology-driven safety solutions.

The convergence of Artificial Intelligence (AI) with immersive technologies such as Augmented Reality (AR) and Virtual Reality (VR) is rapidly transforming the landscape of healthcare and medical training. While AI enhances decision-making through data-driven insights, AR and VR offer intuitive, spatial, and interactive environments that support diagnostics, education, therapy, and surgical planning. However, the integration of these technologies remains fragmented, with varying degrees of adoption, technical maturity, and clinical impact.

This project proposes a systematic literature review to consolidate and critically evaluate current research at the intersection of AI and AR/VR in the medical domain. By synthesising findings across peer-reviewed publications, this review aims to identify common application areas (e.g., diagnosis, surgery, rehabilitation, mental health, etc.), technical integration methods, benefits, and limitations. It will also explore how user interaction, data visualisation, and real-time feedback are managed in immersive AI-enabled medical systems.

Which route is the best to drive from Monash University (Clayton campus) to Melbourne CBD? 

For many of us, answering this question would likely mean opening a route natvigation app and asking the provider to give us the fastest route. For some of us, this question might not need to be answered as you may already be experienced to drive from Monash Uni to CBD, or simply find that the route computed by the app is insufficent to handle your specific requirements, preferences, or constraints. 

Route natvigation is an important tool in driving, and particular useful to drivers who are new or have no experience to the target location. While many drivers would find adequate to use the shortest path (route) by asking the app to solve the simpliest form of natgivation problem - the shortest path problem, many users with complex requirements, preferences, or constraints would find the shortest path solution insufficient to meet their demands. For example:

• My rental vehicle in Europe is pretty large, is the computed path sufficient for the vehicle to pass through?
• My vehicle has low clearance, would the path easy for the vehicle to drive on (with as little patholes/unpaved roads as possible)?
• I am on a sightseeing trip, can we also ensure the route is not boring (with beautiful scenary)? 

Unfortunately, these questions are challenging to be answered by the simple shortest path algorithm. Users often want visual images justifying the choices, explanations on why the path (or a particular road) is selected, and easy to use graphical interfaces to natvigate on all the options. With the current active development of Large Language Model and Chatbot AI, reasoning on routing options and visual images has become feasible.

Are you interested to build a Smart Route Natvigation Recommendation System with us using the latest AI technology? 

In NeuroDistSys (NDS): Optimized Distributed Training and Inference on Large-Scale Distributed Systems, we aim to design and implement cutting-edge techniques to optimize the training and inference of Machine Learning (ML) models across large-scale distributed systems. Leveraging advanced AI and distributed computing strategies, this project focuses on deploying ML models on real-world distributed infrastructures, improving system performance, scalability, and efficiency by optimizing resource usage (e.g., GPUs, CPUs, energy consumption). Researchers and students will explore innovative approaches to reduce latency, increase throughput, and enable real-time resource management, preparing them for impactful roles in AI, cloud computing, and large-scale system design. A practical example of this project includes, but is not limited to, distributed inference of foundation models across heterogeneous server environments.

Feel free to visit my website here and contact me for more information.

In SmartScaleSys (S3), we aim to design and build resource management solutions to learn from usage patterns, predict future needs, and allocate resources to minimize latency, energy consumption, and costs of running diverse applications in large-scale distributed systems. This project offers researchers and students a chance to explore cutting-edge concepts in AI-driven infrastructure management, distributed computing, and energy-aware computing, preparing them for impactful roles in industry and research.

Key Components and Example Scenarios

1. Predictive Resource Allocation and Load Balancing:
   • Example: Imagine an e-commerce platform that experiences a surge in user activity during sales events. SmartScaleSys (S3) would predict this peak demand using ML models and allocate additional resources to avoid downtime, adjusting dynamically as traffic fluctuates.
   • For researchers and students, this component focuses on developing ML models to predict resource needs, improving load distribution to avoid system bottlenecks, and ensuring low-latency performance.
2. Energy-Efficient Operations with Carbon-Aware Scheduling:
   • Example: For non-urgent data processing, SmartScaleSys (S3) could prioritize tasks during off-peak hours when renewable energy sources are more available, reducing carbon emissions.
   • Researchers and students will design policies that align workloads with green energy availability, minimizing energy costs and environmental impact. This offers opportunities to work on practical applications in sustainability for cloud and edge systems.
3. Privacy-Enhancing Resource Management:
   • Example: A healthcare application needs to analyze sensitive patient data across distributed nodes.
   • Researchers and students can explore privacy-preserving algorithms and technologies like federated learning and zero-knowledge proofs, enabling secure, distributed data processing.
4. Self-Healing and Fault Tolerance:
   • Example: If a node in a distributed system fails, SmartScaleSys (S3) would automatically detect this and re-route tasks to maintain uninterrupted service.
   • This research area involves developing fault-tolerant systems that adapt to hardware and software failures. Students will work on predictive maintenance and automated recovery algorithms, improving system resilience.

Research Areas for Master’s and PhD Students

1. AI-Enhanced Resource Forecasting and Optimization:
   • Research Focus: Developing and testing ML algorithms for predicting resource demand and optimizing allocations.
   • Skills Gained: Students will work with real-time data analytics, reinforcement learning, and predictive modeling to create efficient resource utilization strategies.
2. Carbon-Aware and Sustainable Scheduling Algorithms:
   • Research Focus: Creating algorithms that factor in carbon intensity and renewable energy availability to minimize environmental impact.
   • Skills Gained: Students will study sustainability in computing, including integrating AI with real-time energy data and carbon monitoring, contributing to low-impact, sustainable cloud operations.
3. Distributed Federated Learning for Resource Management:
   • Research Focus: Implementing federated learning to optimize resource management across multiple sites without compromising data privacy.
   • Skills Gained: This research offers experience in federated learning, data security, and privacy, preparing students for data-sensitive applications in healthcare, finance, and more.
4. Dynamic Scaling and Load Balancing with Reinforcement Learning:
   • Research Focus: Designing adaptive load-balancing algorithms using reinforcement learning to dynamically respond to changes in workload.
   • Skills Gained: Students will develop expertise in reinforcement learning and its applications in cloud and edge environments, optimizing performance in real-world scenarios.
5. Edge-Cloud Collaboration for Low-Latency Resource Management:
   • Research Focus: Investigating hybrid architectures that balance workload between edge devices and cloud resources to minimize latency.
   • Skills Gained: Students will study distributed systems and hybrid architectures, ideal for IoT applications, real-time data processing, and autonomous systems.

This project will allow students to gain hands-on experience in building, testing, and deploying intelligent resource management tools that not only improve performance but also reduce the environmental footprint of distributed computing systems. By working on SmartScaleSys (S3), students will contribute to the future of sustainable and efficient computing infrastructure.

Feel free to visit my website here and contact me for more information.

This project will involve benchmarking state of the art methods for time series classification on the new MONSTER benchmark datasets [1, 2, 3].  Currently almost all benchmarking in time series classification is performed on the (almost all very small) datasets in the UCR and UEA archives.  This is particularly unsuitable for deep learning models which are low bias models and ideally trained using large quantities of data.  The "true" performance of current deep learning methods for time series classification is unknown outside of the UCR/UEA datasets.  Most deep learning models for times series classification are configured to be trained on tiny quantities of data.  A lot of published work on deep learning for time series classification has serious methodological flaws (e.g., directly or indirectly optimising on the test data).  This project would involve establishing a sound training setup for deep learning models on large quantities of training data, and training and deep learning models for time series classification on the much larger datasets in the new MONSTER benchmark, in order to establish their "true" performance on large datasets.

Politeness plays a pivotal role in fostering constructive, respectful communication and maintaining positive social dynamics in educational settings. In online learning environments—such as discussion forums on Moodle—language becomes the primary medium for interaction between instructors and students. While prior studies have highlighted the benefits of politeness in building rapport and encouraging engagement, limited empirical work has systematically examined how politeness is expressed by both instructors and students in these digital spaces. This project addresses this gap by applying computational techniques to quantify politeness levels in educational forum discussions and analyze how various contextual and social factors influence politeness in instructor-student exchanges.

Are you interested in biomedical? You could combine your data science and computing expertise to analyse DNA and genetics.

Do you play any classical music instruments, like piano or violin? Would you like to combine your advanced music skills with computer science. This project analyses classical music using computer science techniques.
 

Do you play any classical music instruments, like piano or violin? Would you like to combine your advanced music skills with computer science. This project analyses classical music using computer science techniques.

Are you interested in applying your AI/DL knowledge to the medical domain?

Are you interested in programming maps, such as GoogleMaps or Open Street Maps? This project uses online maps extensively for visualising routes, and other objects of interest.

Do you play any classical music instruments, like piano or violin? Would you like to combine your advanced music skills with computer science. This project analyses classical music using computer science techniques.

Are you interested in applying your database knowledge to a real project? This is a collaboration with the Faculty of Medicine, Monash University.

Are you interested in working with hospital data? This project is a collaboration with the Faculty of Medicine, Monash University. In this project, you will be working with medical doctors from Monash Health. 

Large language models (LLMs) have recently made significant progress in machine translation quality [1], but they still struggle with maintaining consistency and accuracy across entire documents. Professional translators commonly use translation memory (TM) tools to reuse past translations, ensuring consistent terminology and phrasing throughout a document. Inspired by the latest research, such as DelTA [2], a document-level translation agent with a multi-level memory architecture, and HiMATE [3], a multi-agent evaluation framework leveraging fine-grained MQM error typology, this project seeks to bridge the gap between LLMs and traditional TM systems. The goal is to enhance domain-specific accuracy and document-level coherence in LLM-based translation by intelligently incorporating a translation memory mechanism.

The research project aims to investigate:

- Multi-Model Fusion with Deep Neural Networks for Future Energy Systems (Smart Grid). 

Future energy systems are envisioned to be running decentrally with full automatic control, high proportion of renewable energy (e.g., wind & solar), and abundant storage facilities. With many types of renewable energy sources are weather and climate dependent, accurate and timely prediction on reliability risks (e.g., loss of generation, voltage issues, and thermal limit violations) due to weather/climate are often necessary.

In power systems, numerical weather/climate data are often chosen to be used for crafting energy system prediction models due to its ease of storage, analysis, and simplicity in training. However, recent research in climate science also shows that a large category of datasets are images (e.g., radar / satellite image dataset), which can also be used to improve prediction for weather/climate-related forecasting. With majority of research in energy/power system focused on numerical weather/climate data, there are huge research potentials in exploring weather/climate data that are not natively numerical.

This project will explore various non-numerical weather/climate data to assist energy reliability prediction, and devise a multi-model fused neural network to effectively capture and predict reliability risks for future energy systems. 

The rapid growth of electric vehicles (EVs) is transforming the transportation systems worldwide. Both EV fleets and private EVs are emerging as a cleaner and more sustainable component of urban mobility, forming an effective way to solve environmental problems and reduce commute costs in future smart cities. Due to the complex spatiotemporal behaviors of passengers and their travel patterns, the unmanaged electric charging demand from EVs may significantly impact the existing transportation and electrical power infrastructure. Without strategic planning and effective operation, the accelerating EV usage could lead to significant challenges to traffic congestion, grid reliability, and suboptimal utilization of resources. Therefore, it is essential to develop reliable, robust, and resilient EV charging networks and charging strategies to address the above-mentioned challenges and facilitate smooth transition to a sustainable urban future.

#sustainability

Thanks to the widespread deployment of smart meters, high volumes of residential load data have been collected and made available to both consumers and utility companies. Smart meter data open up tremendous opportunities, and various analytical techniques have been developed to analyse smart meter data using machine learning. This project will provide a new angle toward energy data analytics and aims to discover the consumption patterns, lifestyle, and behavioural changes of consumers.

This project covers a wide range of research topics in smart meter data analysis toward a better understanding of (electricity and water) consumption behaviours, thus providing insights into the energy programs and policies. For example, previous thesis students have

• explored the human factors driving energy consumption behaviours;
• developed machine learning models to analyse the relationship between users' load patterns and their demographic and socioeconomic factor, promoting energy equity and fairness;
• studied seasonal load patterns and identified essential factors that determine whether users tend to vary their consumption behaviours when the season changes;
• developed machine learning models (including deep learning models) to detect household activities and disaggregate energy consumption for different activities.

There remain many research gaps in the area, and we very much look forward to advancing this research area with you.

#sustainability

Traditional active learning helps reduce labeling costs by selecting the most useful examples from a large pool of unlabeled data. However, in many real-world cases, such a large pool doesn't exist or is expensive to collect. This project explores a new approach using large language models to create synthetic unlabeled text data instead. Rather than just picking data to label, the model will also generate new examples that are diverse and potentially helpful for learning. The aim is to reduce both the amount of data we need to collect and the number of labels required to train accurate models. The project will implement and test different generation strategies to ensure the diversity and coverage of synthetic data, and integrate with active learning to help the model learn faster and better.

This project focuses specifically on LLM applications: chatbots used in customer support (e.g., healthcare). The goal is to investigate how user requirements (e.g., “the bot should de-escalate frustrated users”) can be systematically translated into prompt templates or prompt strategies.

Instead of treating prompt engineering as a trial-and-error process, this project will explore how RE principles, like functional requirements and acceptance criteria, can be mapped to prompt artifacts. The project will evaluate prompt variants and trace their effectiveness in meeting specified requirements using defined metrics (e.g., user satisfaction, task completion rate, tone control).

Keywords: Software Engineering, LLM-based systems, Prompt Engineering, Requirements Analysis.

This project aims to develop privacy-preserving deepfake detection techniques that enable accurate and secure identification of synthetic audio and video content without exposing sensitive user data. Traditional detection methods often require access to raw audio or visual inputs, raising significant privacy concerns, especially in scenarios involving personal or biometric data. Leveraging techniques such as federated learning, differential privacy, and secure multi-party computation, this project seeks to design detection frameworks that maintain high performance while ensuring user data remains decentralized and protected. The outcome will contribute to trustworthy and ethically aligned AI systems that can be deployed in real-world environments, such as social media platforms and communication apps, where privacy and security are paramount.

Mis/disinformation (also known as fake news), in the era of digital communication, poses a significant challenge to society, affecting public opinion, decision-making processes, and even democratic systems. We still know little about the features of this communication, the manipulation techniques employed, and the types of people who are more susceptible to believing this information.

This project extends upon Prof Whitty's work in this field to address one of the issues above.

This project focuses on enhancing security and privacy protection in blockchain-based systems for verifying education credentials, with the goal of combating the proliferation of fake certificates. By leveraging the immutable and decentralized nature of blockchain, the project aims to develop a secure credential verification framework that ensures the authenticity and integrity of academic records while safeguarding users' personal information. The system may incorporate advanced cryptographic techniques, such as decentralized identifiers (DIDs), to enable privacy-preserving verification without compromising trust. Through the design, implementation, and testing of a prototype, the project will explore practical solutions to current challenges in digital credentialing, contributing to increased trust and transparency in educational and employment processes.

Background and motivation

As intelligent agents make decisions, any project aiming to realize human-like AGI should model decision-making.  As we have been pursuing the WBA approach to create AGI by learning from the architecture of the entire brain, we request you to model the decision-making of the mammalian brain.  While a number of models have been proposed, we refer to O’Reilly’s model on his textbook for computational cognitive neuroscience (CCNBook hereafter) as the standard, where decisions are supposed to be made with the loop consisting of the frontal cortex, basal ganglia, and related areas, which reinforces decisions in an actor-critic way.

Objective

You are requested to implement a biologically-plausible yet computationally effective model for decision-making and action selection.  The model should serve as a reference model for other brain-inspired models of intelligence. Thus, its implementation should be as simple as possible for being used and maintained in the community.  We outline the structure of the model to be implemented in the Detailed Project Description section below.

Success criteria

The implementation will be judged with the following criteria:

• Biological plausibility
  The implementation should be ‘compatible’ with the structure and function of the mammalian brain.
• Usability
  The implementation should be easily used and maintained together with documentation.
• Specifications
  See the Detailed Project Description section below.
• Performance
  Use one or more tasks from the Dataset and Test section below.

Detailed Project Description

The request is to implement a decision-making model consisting of the following modules: (They are refactored from the model in the CCNBook.)

• FC Module
  • provides with options based on input from outside
  • may or may not implement a winner-take-all logic for decision-making
  • may or may not implement an accumulator logic to accumulate scores for options
  • may use recurrent networks
• Actor module
  • corresponds to part of BG and the thalamus
  • modulates the strength of each option
  • receives the TD signal from the Critic to learn
  • receives the state input from outside
  • may or may not implement a winner-take-all logic for decision-making
• Critic module
  • corresponds to part of BG and the amygdala
  • creates the TD signal based on the exterior reward
  • receives the state input from outside

Model characteristics:

The model should have the following performance characteristics (see here for a discussion of these constraints):

• Few hyperparameters: These should either adapt automatically to model conditions, or not vary over time.
• Possibility to select 1 or more actions simultaneously, if necessary
• Conflict resolution: A method of excluding incompatible action selections
• Clean switching: Marginally better actions should be selected quickly and definitively without vacillation or dithering.
• Full selection: Options that were not selected should not interfere the selected action.

Dataset and Tests

While currently we do not offer our own dataset or test batteries, the Executive Function chapter of CCNBook and the experiment library at psytoolkit.org refer to tests for decision-making models.  You can choose one or more tasks here to test your implementation.  Note that most of them require working memory so that they test working memory as well.

• Task switching tasks (psytoolkit.org):

• • Wisconsin card sorting task (WCST)
  • Dimensional change card sorting task (DCCS)
  • N-Back task (Wikipedia)

• Stroop task (CCNBook) / (Psytoolkit.org) (Wikipedia)
• Store-Ignore-Recall (SIR) task (CCNBook)
• Reproducing the A Not B error (CCNBook) (Wikipedia)

Markov Decision Processes (MDPs) are frameworks used to model decision-making in situations where outcomes are partly random and partly under the control of a decision maker. While small MDPs are inherently interpretable for people, MDPs with thousands of states are difficult to understand by humans. The K-MDP problem is the problem of finding the best MDP with, at most, K states by leveraging state abstraction approaches to aggregate states into sub-groups. The aim of this project is to measure and improve the interpretability of K-MDP approaches using state-of-the-art XAI approaches. We will instantiate and evaluate our approaches on a range of computational sustainability case studies from the domain of conservation of biodiversity, natural resource management and behavioural ecology. 

This is a Winter Student Research Internship ONLY not an honours or minor thesis project at this time.

Please apply here if you are interested in the role before the deadline: 

https://www.monash.edu/study/fees-scholarships/scholarships/summer-winter

__________

We're reimagining the Control Room of the Future, where advanced data tools support better decision-making in complex environments like energy grid operations. A key focus of our research is understanding how operators interact visually with large-scale information displays. To do this, we use eye-tracking technology to capture detailed visual attention patterns, and synchronise it with workstation-level video and researcher notes. 

About Gazealytics

We use Gazealytics to support the analysis of eye-tracking data in context. Gazealytics is an open-source visual analytics toolkit for eye-tracking research, available here: https://github.com/gazealytics/gazealytics-master. It can help researchers synchronise and analyse eye-tracking data alongside other data streams such as video recordings and observer notes.

Your Role

You will help to extend Gazealytics' capabilities for more advanced visual and temporal analysis. Key development tasks may include:

• Implement temporal Area of Interest (AOI) tracking
• Develop visual overlays of gaze data on synchronised video
• Enhance the timeline-based data synchronisation (eye-tracking, video, notes)
• Optimise for performance and scalability with large video datasets
• Improve the frontend UX for easier data navigation and filtering

Tech Stack

• Frontend: JavaScript (with D3.js for visualisation), HTML5, CSS3
• Backend: Python (Flask)
• Data Handling: JSON-based data structures, TSV imports, temporal data synchronisation
• Version Control: GitHub

Goal recognition is defined as the problem of determining an agent’s intent from observations of its behaviour. Current research in goal recognition has focused on observing agents that are trying to achieve their goals in a rational manner. Other research has focused on observing agents that are deliberately trying to trick an observer into believing they are pursuing alternative goals to the ones they are actually pursuing. However there is also a need to recognise when a behaviour is suspicious, regardless of the goal that is being tried to be achieved. This project focuses on the generation of suspicious behaviours according to novel models.

Classical propositional logic (CPL) captures our basic understanding of the linguistic connectives “and”, “or” and “not”. It also provides a very good basis for digital circuits. But it does not account for more sophisticated linguistic notions such as “always”, “possibly”, “believed” or “knows”. Philosophers therefore invented many different non-classical logics which extend CPL with further operators for these notions.

Over the past fifty years, such non-classical logics have proved vital in computer science and logic-based artificial
intelligence: after all, any intelligent agent must be able to reason with more than just and-gates, not-gates and or-gates! For example, we now have non-classical logics which capture notions such as “phi is true until psi becomes true”; “if we execute atomic program alpha in state x then we end up in a state y which obeys formula psi”, and many more.

The most common way to determine the true or false status of a given formula is to use the tableau method [1]. We have developed and implemented such methods for a plethora of non-classical logics [2]. But how can we guarantee that the implementation is faithful to the theory? Indeed, how can we be sure that we have not made a mistake in the theory?

Over the past twenty years, we have shown that we can encode the theory of the tableau method into an interactive proof-assistant, such as Coq [3]. The proof-assistant checks all of the claims we make about our tableau rules. We then interact with the proof-assistant to prove the following claim: for all formulae phi, it is possible to decide whether phi is true or false in the given non-classical logic.

The proof of the claim contains an algorithm for deciding whether an arbitrary formula is true or else false! This proof can then be exported automatically to produce a formally verified computer program that implements the decision procedure [4].

Your project is to continue this work and to hopefully publish an academic paper in an international conference.

You will need a strong background in maths but no programming skills in languages such as C++. You will need to read and understand the previous work, and then learn to program in Coq. The research is to extend the existing work to new logics.

[1] Rajeev Gore: Tableaux Methods for Modal and Temporal Logics. Handbook of Tableau Methods, Kluwer, 1999.

[2] Rajeev Gore, Florian Widmann: Optimal and Cut-Free Tableaux for Propositional Dynamic Logic with Converse. IJCAR 2010:
225-239.

[3] Jeremy E. Dawson, Rajeev Gore: Generic Methods for Formalising Sequent Calculi Applied to Provability Logic. LPAR 2010:
263-277.

[4] Minchao Wu, Rajeev Gore: Verified Decision Procedures for Modal Logics. International Conference on Interactive Theorem
Proving 2019: 31:1-31:19.

Classical propositional logic (CPL) captures our basic understanding of the linguistic connectives “and”, “or” and “not”. It also provides a very good basis for digital circuits. But it does not account for more sophisticated linguistic notions such as “always”, “possibly”, “believed” or “knows”. Philosophers therefore invented many different non-classical logics which extend CPL with further operators for these notions.

Over the past fifty years, such non-classical logics have proved vital in computer science and logic-based artificial
intelligence: after all, any intelligent agent must be able to reason with more than just and-gates, not-gates and or-gates! For example, we now have non-classical logics which capture notions such as “phi is true until psi becomes true”; “if we execute atomic program alpha in state x then we end up in a state y which obeys formula psi”, and many more.

Of course, there is no such thing as a free lunch, for it is significantly harder to decide whether a non-classical formula is true or false (EXPTIME vs NP). Can we develop and implement efficient algorithms for this problem?

This problem has been attacked using multiple different methods for the past 40 years, without much success. But, in 2021, second-year student Cormac Kikkert and I showed that we could improve the existing decision procedures for these logics by orders of magnitude by combining two well-known existing methods giving rise to what we called CEGAR-tableaux [1].

Your project is to continue this work and to hopefully publish an academic paper in an international conference.

You will need excellent programming skills and a good background in maths. This project would set you up for a follow-up honours project in this area.

https://github.com/cormackikkert/CEGARBox

https://github.com/cormackikkert/CEGARBoxCPP

Rajeev Goré, Cormac Kikkert: CEGAR-Tableaux: Improved Modal Satisfiability via Modal Clause-Learning and SAT. TABLEAUX 2021: 74-91.

Is the user paying attention? Is the content engaging enough?

The degree of concentration, enthusiasm, optimism, and passion displayed by individual(s) while interacting with a
machine is referred to as ‘user engagement’. Engagement is a positive psychological state characterized by active behavioral participation, positive emotional experiences, and intense cognitive focus. Being able to detect engagement and/or attention has wide applications in consumer commerce, smart cars, augmented reality etc. 

This project deals with developing multimodal neural networks for predicting engagement and attention of user.

Is the user paying attention? Is the content engaging enough?

The degree of concentration, enthusiasm, optimism, and passion displayed by individual(s) while interacting with a
machine is referred to as ‘user engagement’. Engagement is a positive psychological state characterized by active behavioral participation, positive emotional experiences, and intense cognitive focus. Being able to detect engagement and/or attention has wide applications in consumer commerce, smart cars, augmented reality etc. 

This project deals with developing multimodal neural networks for predicting engagement and attention of user.

Deepfakes detection deals with machine learning methods, which detect if an image/video/audio sample is manipulated with a generative AI software. In recent years, deepfakes have been increasingly used for malicious purposes, including financial fraud, misinformation campaigns, identity theft, and cyber harassment. The ability to generate highly realistic synthetic content poses a serious threat to digital security, privacy, and trust in media. This project will develop methods for detecting deepfakes.

The world’s energy markets are transforming, and more renewable energy is integrated into the electric energy market. The intermittent renewable supply leads to unexpected demand-supply mismatches and results in highly fluctuating energy prices. Energy arbitrage aims to strategically operate energy devices to leverage the temporal price spread to smooth out the price differences in the market, which also generates some revenue. The ancillary energy market provides frequency regulation services to ensure power system security in the face of limited dispatchability of the renewable supply. It is often difficult to forecast prices in the energy market and challenging to develop a bidding strategy for arbitrage, given the market's complex behavior.

This project aim to design effective bidding strategies to leverage historical market data and secure market operation. Deep reinforcement learning is a category of artificial intelligence that learns the best actions through a series of trials and errors similar to humans. Deep reinforcement learning's unique features make it an ideal technology for dynamic and stochastic environments, such as energy markets. But most existing studies focused on either a single market using reinforcement learning or multiple markets using optimisation methods given market prices, leaving a research gap of how to bid in multiple markets in real-time optimally.

#sustainability

Computational modelling of sign languages aims to develop technologies that can serve as means for understanding (i.e., recognising) and producing (i.e., generating) a particular sign language. The objective of this project is to study and contribute to the state-of-the-art in sign language segmentation.

This is a research project best for students who are independent and willing to take up challenges. It is also a good practice for students who wish to pursue further study at a postgraduate level.

Project Description:

The project is focused on developing game-based learning environments where the users’ trace or interaction data could be collected. The game-based environment needs to be designed to allow the users to navigate and explore at their own pace. Using the environment, the participants can practice their technical/professional skills from various options.

The game design and development needs to be completed in Semester 1. Semester 2 is where the game will be tested on participants, and data will be collected. This project will offer a way to assess how users’ skills develop and what parts of the game or user behaviours are more critical for their skill acquisition and development.

Technical and research guidance would be provided. This experience can offer you an opportunity to familiarise yourself with research-based design.

Project description

This research project aims to explore how researchers are investigating various epistemic emotions that arise within educational contexts. Through a combination of literature review, citation and bibliometric analysis, this project will uncover trends in the field and may guide future directions.

The outcome of this research is expected to be a peer-reviewed publication that may provide a valuable addition to your CV.

Mental health challenges disproportionately affect vulnerable populations, often due to limited access to traditional healthcare services. The rise of Generative AI offers a groundbreaking opportunity to bridge this gap by providing personalized, scalable, and accessible mental health support. This project, led out of Action Lab, aims to harness the potential of Generative AI to develop innovative technologies tailored for mental health interventions.

About us

Action Lab is multidisciplinary team of 30 impact-focused researchers, designers and engineers working at the intersection of communities, technology and social innovation. Systems designed and built at Action Lab (always in collaboration with real-world organisations and academic partners) are used nationally and internationally.

Roisin McNaney and Delvin Varghese are academics based at Action Lab in the Department of Human-Centred Computing. Roisin leads the CSIRO Next Gen Graduate Program in AI & Mental Health.

Computational modelling of sign languages aims to develop technologies that can serve as means for understanding (i.e., recognising) and producing (i.e., generating) a particular sign language. The objective of this project is to study and contribute to the state-of-the-art in sign language recognition.

This is a research project best for students who are independent and willing to take up challenges. It is also a good practice for students who wish to pursue further study at a postgraduate level.

Computational modelling of sign languages aims to develop technologies that can serve as means for understanding (i.e., recognising) and producing (i.e., generating) a particular sign language. The objective of this project is to study and contribute to the state-of-the-art in sign language generation.

This is a research project best for students who are independent and willing to take up challenges. It is also a good practice for students who wish to pursue further study at a postgraduate level.

Recognising conversational nonverbal behaviour for speakers and listeners, such as hand gestures, facial expressions and eye-gaze, is of great importance for natural interaction with intelligent agents. The objective of this project is to study and contribute to the state-of-the-art in conversational nonverbal behaviour recognition.

This is a research project best for students who are independent and willing to take up challenges. It is also a good practice for students who wish to pursue further study at a postgraduate level.

Generating conversational nonverbal behaviour for speakers and listeners, such as hand gestures, facial expressions and eye-gaze, is of great importance for natural interaction with intelligent agents. The objective of this project is to study and contribute to the state-of-the-art in conversational nonverbal behaviour generation.

This is a research project best for students who are independent and willing to take up challenges. It is also a good practice for students who wish to pursue further study at a postgraduate level.

Our groundbreaking research explores the intricate relationship between natural language processing (NLP) and electroencephalography (EEG) brain signals [1]. By leveraging advanced machine learning techniques, we aim to decode the neural patterns associated with language comprehension and production, ultimately enabling seamless communication between humans and machines. Our innovative approach has the potential to revolutionize brain-computer interfaces,speech recognition technologies, and assistive devices for individuals with communication impairments.

[1] Mohammadi Foumani, N., Mackellar, G., Ghane, S., Irtza, S., Nguyen, N., & Salehi, M. (2024, August). Eeg2rep: enhancing self-supervised EEG representation through informative masked inputs. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (pp. 5544-5555).

Collecting and analysing social media content (e.g., Reddit), along with using Google Trends, presents a great opportunity to develop social media epidemic intelligence. This approach can enhance the understanding of chronic conditions such as arthritis, back pain, and knee pain, as well as track associated areas such as treatments and risk factors, including obesity, diet, physical activity, and exercise. This approach can be also used to understand the community attitudes about these conditions, and see if there are changes over time as there as new public campaigns.

Current studies on diabetes recommender systems and apps mainly focus on the performance and personalisation of AI models and techniques, including machine learning and deep learning models that are trained on user data. These works often use a one-size-fits-all approach for presenting information to users. Yet, research shows that humans process information in different ways, and their attitudes towards an action depend on their attitude-function styles. The success of recommender apps for diabetes management to effectively influence behaviour heavily relies on considering social aspects of human behaviour, which have been largely overlooked in the literature. This represents a significant knowledge gap in this research area.

Current studies on diabetes recommender systems and apps mainly focus on the performance and personalisation of AI models and techniques, including machine learning and deep learning models that are trained on user data. These works often use a one-size-fits-all approach for presenting information to users. Yet, research shows that humans process information in different ways, and their attitudes towards an action depend on their attitude-function styles. The success of recommender apps for diabetes management to effectively influence behaviour heavily relies on considering social aspects of human behaviour, which have been largely overlooked in the literature. This represents a significant knowledge gap in this research area.

This project aims to develop effective machine learning algorithms for detecting deepfake videos, which have become a significant concern for disinformation and cybersecurity. The objectives include pre-processing the data for feature extraction, and training machine learning models to accurately classify videos as either real or manipulated. The methodology involves using advanced techniques such as convolutional neural networks, recurrent neural networks or video vision transformer models to analyse visual and temporal patterns in the videos. In addition, techniques like facial recognition, frame analysis, and optical flow detection may also be employed to identify inconsistencies or artifacts typical of deepfake generation. These techniques may help to enhance the video deepfake detection performance.

This project aims to develop foundation models for detecting anomalies in time series data. Anomalies, such as unusual patterns or unexpected events, can signal critical issues in systems like healthcare, finance, or cybersecurity. Current methods are often limited by the fact that they reuire long training before one can test the model on a new time series due to complexity and variability of real-world time series data. By leveraging advanced machine learning techniques, this project seeks to create robust and adaptable models that can generalize across diverse time series scenarios. The expected outcomes include improved accuracy in detecting anomalies. The benefits span various sectors, including cybersecurity and healthcare.

Background:

Imagine the human brain as a complex electrical grid, with over 80 billion neurons (nerve cells) acting as power stations. These power stations need to send electrical signals to each other efficiently. Myelin, a special lipid sheath, wraps around the neuron processes (axons) like insulation around electrical wires. This insulation ensures that the signals travel quickly and without losing strength, giving the brain’s “white matter” its name (Figure 1A).

When diseases like multiple sclerosis strike, it is as if the insulation around the wires gets damaged, causing the electrical signals to slow down. Over time, this damage can also affect the wires themselves, leading to even more significant problems, much like an electrical grid experiencing power outages.

To understand how white matter develops and is affected by disease, we use advanced imaging techniques like Magnetic Resonance Imaging (MRI) (Figure 1A). By utilising AI and deep learning, we can detect subtle changes in white matter volume. This is akin to having a sophisticated system that can monitor the health of the electrical grid, detecting even the smallest issues in the insulation, and providing valuable insights into the development and progression of white matter diseases.

This project focuses on identifying and distinguishing between authentic audio recordings and those that have been artificially generated or manipulated. As voice cloning technology advances, creating realistic audio deepfakes has become easier, raising concerns about misinformation and privacy. To combat this, this project aims to develop machine learning models to analyse audio features such as pitch, tone, cadence, and spectral characteristics. These techniques are implemented to detect subtle anomalies that may indicate manipulation, even in high-quality deepfake audio. Additionally, training on large and diverse datasets that include both genuine and synthetic voices allows these models to improve their accuracy and robustness. Apart from proposing and implementing deep learning methods for detection purposes, this project also involves collecting and/or generating different datasets for experiments.

This project aims to identify the geographical position where an audio clip was recorded by analysing sound patterns and audio signals from the surrounding environment. This approach leverages hand-crafted and/or deep features to distinguish between different soundscapes associated with specific locations, like train stations, shopping malls, classrooms, hospitals, parks, and so on. Deep learning models are trained on labelled audio datasets that capture diverse environments and their unique acoustic signatures. By processing audio input, these systems can infer location based on the recognized sounds, such as traffic noise, voices, or ambient sounds. The project includes gathering datasets of audio clips with known geographical locations to train the machine learning models.

This project involves enhancing traditional object detection methods by incorporating human pose estimation to identify weapons in various contexts, especially in surveillance and security applications. This approach leverages computer vision techniques that analyse the positions and movements of individuals, allowing systems to recognize not just the presence of weapons but also the intent and behaviour of the person carrying them. By integrating pose data with advanced machine learning methods, the system can more accurately recognise threatening situations, distinguishing between benign gestures and potential threats. This analysis improves detection accuracy in complex environments, such as crowded public spaces. The project may cover different types of weapons, but it will primarily focus on two major ones including guns and knives.

This project aims to develop techniques that enable users to find relevant audio content by inputting textual queries. This process leverages machine learning models, particularly natural language processing and audio signal processing, to bridge the gap between text and audio. When a user submits a query, the system analyses the text to understand its intent and context. It then searches a database of audio files, employing techniques such as keyword extraction, semantic understanding, and even speech recognition, to match the query with relevant audio clips. Recent state-of-the-art deep learning methods will be thoroughly reviewed to identify their strengths and weaknesses. Additionally, these methods will undergo empirical evaluation to assess their performance in practical applications, providing insights into their effectiveness and potential improvements. These approaches enhance the efficiency of locating specific sounds, speeches, or music within large collections, making it especially useful in many applications.

This project involves the automated generation of textual descriptions for audio content, such as spoken language, sound events, or music. This process typically employs deep learning techniques, such as recurrent neural networks, transformer models, and so on, to analyse audio signals and generate coherent captions. By training on large datasets that include both audio recordings and corresponding textual descriptions, these models learn to recognize patterns and contextual meanings within the audio. This project entails the collection and generation of audio-description datasets to create a robust foundation for analysis. In addition, various deep learning models will be proposed and implemented to explore their effectiveness in processing and interpreting the audio data. Finally, a comprehensive evaluation will be conducted to assess the performance of these models, identifying their strengths and areas for improvement.

Adaptively smoothing one-dimensional signals remains an important problem, with applications in time series analysis, additive modelling and forecasting. The trend filter provides an novel class of adaptive smoothers; however, it is usually implemented in a frequentist framework using tools like the lasso and cross-validation. Bayesian implementations tend to rely on posterior sampling and as such do not provide simple, sparse point-estimates of the underlying curve.

Learning appropriate prior distributions from replications of experiments is a important problem in the space of hierarchical and empirical Bayes. In this problem, we exploit the fact that we have multiple repeats of similar experiments and pool these to learn an appropriate prior distribution for the unknown parameters of this set of problems. Standard solutions to this type of problem tend to be of mixed Bayesian and non-Bayesian form, and are somewhat ad-hoc in nature.

The spectral density of a time series (a series of time ordered data points -- for example, daily rainfall in the Amazon or the monthly stocks of fish in the Pacific) gives substantial information about the periodic patterns hidden in the data. Learning a good model of the spectral density is usually done through parametric methods like autoregressive moving average processes [1] because non-parametric methods struggle to deal with the interesting “non-smooth” nature of spectral densities. This project aims to apply a powerful and new non-parametric smoothing technique to this problem.

Feedback is crucial to learning success; yet, higher education continues to struggle with effective feedback processes. It is important to recognise that feedback as a process requires both teachers and students to take active roles and work as partners. However, one challenge to facilitate a two-way process of feedback is the difficulty to track feedback impact on learning, particularly how students interact with feedback. This project seeks to address this problem by taking a human-centred approach by involving students and teachers to design and implement a learning analytics-based feedback tool to collect and analyse data generated when students make sense of the feedback that they receive. 

Verifiable Dynamic Searchable Symmetric Encryption (VDSSE) enables users to securely outsource databases (document sets) to cloud servers and perform searches and updates. The verifiability property prevents users from accepting incorrect search results returned by a malicious server. However, the community currently only focuses on preventing malicious behavior from the server but ignores incorrect updates from the client, which are very likely to happen in multi-user settings. Indeed most existing VDSSE schemes are not sufficient to tolerate incorrect updates from users. For instance, deleting a nonexistent keyword-identifier pair can break their correctness and soundness. 
 

The increasing integration of Large Language Models (LLMs) into various sectors has recently brought to light the pressing need to align these models with human preferences and implement safeguards against the generation of inappropriate content. This challenge stems from both ethical considerations and practical demands for responsible AI usage. Ethically, there is a growing recognition that the outputs of LLMs must align with laws, societal values, and norms. Practically, the widespread application of LLMs in sensitive contexts necessitates robust measures to prevent the misuse of technology, including content that may circumvent designer-imposed restrictions. 

Researchers highlight the ethical risks in the potential leakage of sensitive medical data inputted into LLMs and the bias in the generated contents. The generative capacity of the LLMs can be misused to disseminate misinformation and disinformation on a massive scale. Despite collective efforts to curtail the generation of undesirable content, innovative misconduct such as prompt injection and other forms of misuse remain prevalent. This scenario underscores the urgency of developing more resilient strategies to harness the transformative potential of LLMs while navigating the complex terrain of safety, ethics, and compliance.

Existing methods focus on actively guiding models to converge to human preference by fine-tuning the model using datasets aligned with human preference and adopting a reward-guided learning procedure. Nonetheless, it is difficult to evaluate the effectiveness of these methods when it comes to predicting the output of LLMs in complex real-world applications with multi-shot questions and constant iterations of model architectures and fine-tuning.

Graph neural networks (GNNs) are widely used in many applications. Their training graph data and the model itself are considered sensitive and face growing privacy threats.
 

With the glow of digital information techniques, mobile systems are powerful ever and occupying more market shares. Just like wildly used social media sites, e.g. Facebook and Twitter, smartphone usage is up to 80% by 2020. In parallel to this trend, many companies are trying to incorporate Artificial Intelligent especially deep learning empowered applications into devices to further ease the life of people. At an early stage, the models are offloaded to the cloud and require frequent data transfer to perform inference since in the mobile end the high computational consumption and complex structure of deep learning models are not affordable. In recent years they are trying to lighten the deep learning models to enable inference and even training at mobile devices directly. Techniques like pruning, quantizing, knowledge distillation are developed to delimitate computation labour of computer vision (CV) tasks. In addition to the help of a new mobile end, deep learning framework (Tensorflow lite, PytorchMobile, CoreML) makes the market of on-device applications getting bigger and bigger due to the low latency and better privacy protection for individuals. 

The last several years have witnessed the promising growth of AI-empowered techniques in mobile devices, from the camera to smart assistants. Users can find traces of AI in almost every aspect of mobile devices. The global mobile artificial intelligence market has reached $8.56 billion in 2020 and is expected to have a growth rate of 6.44% from 2021 to 2030. As the technology keeps advancing, energy-friendly and low latency on-device AI solutions can be the entrance to the next level of development and innovation in the field like AR and autonomous driving and reducing the reliance on cloud AI operations. And using AI to contextualise user behaviours into applications will make each app session more valuable than the last. Compared with the previous ways of selling technology products, delivering compelling and personalised experiences and services has become an essential part of vendor roadmaps for upcoming years.

Based on this trend, as approaching the end of 2021, one of the most widely used deep learning frameworks Tensorflow has enabled its new feature of enabling on-device training in the Tensorflow Lite. Models can be easily finetuned locally to achieve better flexibility and personalization to users. However, on the other side, it also brings the question of the security and privacy issues of deep learning models to a new level and demand. In the past serval years, scholars have put attention on security issues for deep learning models like trojan attacks, information leakage, or adversarial attacks, and have come up with different ways of defending and preserving the robustness of deep learning models in inference time. It is also essential to get a better understanding of how reliable on current on-device training schema is and pave the way for further application of on-device AI models.

Electricity is an essential part of modern life and the economy. Driven by a combination of policy support and rapidly falling costs of low-carbon technologies, Australia is experiencing a sharp rise in the deployment of distributed energy sources (DERs). Typical DERs include wind, solar photovoltaics (PV), battery storage, and electric vehicles (EVs) on the consumer side.

DERs on the consumer side are becoming smart and able to respond to the system variables, such as prices and availability of sustainable energy generation. Such interactions between the grid and consumers are not captured in the current energy system planning and operation. Effective modelling of DER behaviour in the energy system is a key step for capturing its impact on the energy system operation and planning. An effective approximation is needed to capture such aggregate DER behaviours, and deep learning is a promising solution.

#sustainability

Propositional satisfiability (SAT) is a well-known example of NP-complete problems. Although NP-completeness may be perceived as a drawback, it allows one to solve all the other problems in NP by reducing them to SAT and relying on the power of modern SAT solvers. This is confirmed by a wealth of successful examples of use cases for modern SAT solving, including generalisations and extensions of SAT as well as a wide variety of practical applications in artificial intelligence (AI). This project will develop various ways to improve PySAT,  a critical Python Package Index (PyPI) package used for effective prototyping and problem solving with SAT, and augment PySAT with additional functionality.

Online fraud, also referred to as cyberscams, is increasingly becoming a cybersecurity problem that technical cybersecurity specialists are unable to effectively detect. Given the difficulty in the automatic detection of scams, the onus is often pushed back to humans to detect. Gamification and awareness campaigns are regularly researched and implemented in workplaces to prevent people from being tricked by scams, which may lead to identity theft or conning individuals out of money. Whilst these interventions hold some promise they often fail due to ignoring human behaviours and predictors of vulnerability. Arguably, cybersecurity scholars need to learn much more about the methods criminals employ to trick victims out of money.

This honours theiss will  support a project that considers one of the following issues:

- Victimology of cyberscam victims

- Innoculation to cyberscams

- Examination of victims' cyberscam journeys

- Effective methods to change behavious to prevent cyberscam victimisation

Using Adobe Photoshop to generate or edit images is old news, we don't need to edit using our hand and mouse, nor even a stylus pen.  All we need to do is command the gen AI to do it.

Right now the AI literature is trending with many techniques that enable to generate or edit realistic pictures by simply describing them: let there be pictures.

Gen AI has taken the world by storm, it's been applied to many disciplines including in pure sciences.  Notably, Google Deepmind used their graph deep learning based generative AI model (GNoME) to discover millions of new materials, as well as their AlphaFold to predict the structure & interactions of all of life’s molecules.

The energy transition to net zero is in full swing! We at Monash University's Faculty of Information Technology (FIT) are in the unique position that we support the transition across an immensely broad range of topics: from model-predictive building control and community battery integration to wind farm optimisation and multi-decade investment planning, we support clever algorithms and data crunching to make decisions automatically and to let humans make informed decisions, too. 

Across FIT, we have at least one dozen academics working in the greater space of "energy" (optimisation, modelling, software, human-computer interaction, ...). We also work very much interdisciplinarily with colleagues from other faculties, e.g. on bio-diversity matters, on physical aspects, on modelling aspects, and on market design!

If this sounds like it may be of interest, get in touch with Markus, who acts as a facilitator across FIT and who will put you in touch with the right people then.

While our websites are being set up, those from the Monash Energy Institute will give you a first idea of the breadth of our activities: https://www.monash.edu/energy-institute

Note: while only "Optimisation" is selected as the Research Area, please contact us if *anything* in the greater space of the net zero energy transition if of interest to you.

Relevant FIT colleagues include, in no particular order:
Sarah Goodwin, Hao Wang, Terrence Mak, Yolande Strengers, Sarah Pink, Maria Garcia de la Banda, John Grundy, Pierre Le Bodic, Peter Stuckey, Carsten Rudolph, Kari Dahlgren, ... and myself (Markus Wagner)

Genetic improvement (GI) of software is a family of techniques that can automatically improve code using evolutionary algorithms. The idea is to apply changes (swap lines/blocks, change + to -, etc.) to existing code until it is improved. This has been successfully deployed for automated bug fixing, speeding up existing programs, and making software more energy efficient, with impressive results.

Gin (https://github.com/gintool/gin) is an open source tool for GI in Java, and it has been used to support many researchers investigating GI. However, there are ***several*** aspects of Gin that need improved to support modern and large-scale software projects. Several aspects of Gin's integration with the Maven and Gradle build tools needs brought up to date including automatic generation of unit tests and configuration of the profiling tool. Gin's profiling tool was recently updated to support modern Java, but lost some functionality that needs reintroduced. Reporting of exceptions and errors in the software being improved is minimal, but could be added to help both users of Gin, and to provide feedback for Gin's large language model integration. A more ambitious project could also extend to allowing Gin to alter multiple class files at once, or extend the existing basic LLM integration. This project is a challenging combination of problem solving and working in a cutting edge research area.

Islam has become the current fastest growing religion in the world. It indicates that a lot of people starting to get interested to study Islam. The best way to study a religion is by reading its holy scripture, where in Islam is the Quran. People who embrace Islam as their religion are called Muslims, and according to them it requires to sincerely open your mind and heart to study Quran, and it will take a lifetime to understand and analyze the deep meaning of Quran. Therefore, most Muslims study Quran with guidance from Muslim scholars to get better understanding of what they read. However, sometimes people find it difficult to learn and memorize the verse in Quran and also the meaning of it. This minor thesis is a project to develop NLP based tool to analyze the learning process of memorizing the Quran.

Despite the needs of stakeholders, which dental institutions as primary focus in this paper, the feasibility of integrating artificial intelligence (AI), specifically Language Models (LLMs), whether in a broad context or within specific environments such as dental education, is an active area of research. Studies related to AI applications, both in education or dental health institutions, suggest that ChatGPT's performance needs to be considered from various perspectives. While ChatGPT is utilized across diverse subject domains by various users, previous studies have revealed variations in its performance quality. Notably, ChatGPT excels in critical and higher-order thinking tasks within fields like economics, but its performance may not be as satisfactory in domains such as law, mathematics, and medical education. This review builds upon previous research on the use of AI, specifically ChatGPT, within dental education. It investigates several key areas including ChatGPT's capacity to match the assessments of human experts on dental education tasks. This study aims to explore how AI/LLMs could become practical tools in dental health, a field impacting many people. A deeper understanding of these aspects could provide insights relevant to dental education and even broader applications like direct patient interactions.

The integration of chatbots into customer service operations has undergone a transformative shift across diverse industries, becoming essential instruments for delivering assistance and personalized interactions. This development, spurred by the requirement for efficient, scalable, and 24/7 available customer service solutions, has become notable in various sectors, from retail to healthcare. In the electricity services sector, especially in Indonesia, electricity is not just a service but a vital necessity that needs to be available 24/7 to ensure the functioning of various aspects of daily life. Many aspects of life depend on the availability of electricity. This is where the role of customer service becomes crucial as a bridge to fulfill the needs of customers. One common example is reporting power outages. With the presence of customer service, it can help expedite the handling of electrical disruptions. This raises concerns for the efficiency and responsiveness of customer service in addressing these critical issues. This research project specifically addresses the enhancement of chatbot capabilities within the electricity services sector, drawing insights from various studies.

Word sense disambiguation (WSD), the process of computationally identifying the appropriate meaning of a word within its context, is a fundamental task in Natural Language Processing (NLP). Effective WSD is crucial for building accurate machine translation systems, information retrieval tools, and sentiment analysis applications, especially when dealing with diverse languages and linguistic variations. While word sense disambiguation research has made substantial progress for resource-rich languages like English, its application to low-resource languages like Indonesian presents a significant opportunity for further development and exploration. In this project, we will develop several WSD benchmark datasets for low-resource Indonesian local languages.

The Javanese language, spoken by a population of over 98 million people, faces notable challenges in digital and technological applications, especially when compared to globally recognized languages. This disparity is highlighted in several studies that discuss the lack of deep learning research benefits due to data scarcity for Javanese. Additionally, other studies have pointed out the inaccessibility of data resources and benchmarks for Javanese, contrasting with languages like English and Mandarin Chinese. They further emphasize the under-representation and low-resource nature of Javanese in Natural Language Processing (NLP) research. These studies collectively underscore the urgent need to improve the digital and technological infrastructure for the Javanese language. One interesting aspect of the Javanese language that makes it a bit harder to generate language technology for it is its honorific system. Embedded in the social and cultural fabric of Javanese society, the honorific system in Javanese uses different levels of speech to show respect and social hierarchy. These complexities represent both a challenge for NLP research and an opportunity to better understand and model the complexity within language, culture, and social structure. The primary objective of this research is the development of a comprehensive Javanese corpus, with a special focus on its complex honorific system. This corpus aims to provide a rich, detailed dataset that can be utilized to enhance NLP classification tasks, such as sentiment analysis, topic classification, and machine translation, specifically tailored for the Javanese language. By capturing the nuances and variations inherent in the honorific system, the corpus will facilitate the creation of more accurate and culturally sensitive computational models. This will contribute significantly to the field of NLP by providing insights and methodologies that can be applied to other low-resource languages with similar linguistic features.

Political polarization is a phenomenon that permeates societies worldwide, manifesting in divergent ideologies, entrenched viewpoints, and societal fragmentation. In the context of Indonesia, a diverse and populous nation with a complex socio-political landscape, understanding political polarization is crucial for fostering social cohesion and effective governance. However, despite the evident presence of political polarization within Indonesian society, the absence of a comprehensive political polarization corpus poses a significant impediment to the systematic analysis and understanding of this phenomenon. In this project, we will develop a corpus of social media posts annotated for polarization and analyzed based on political identities.

In the evolving landscape of data reporting, traditional text-based and quantitative methods are increasingly being supplemented by rich, community-generated qualitative data, including audio and video content. This shift presents unique challenges and opportunities in how non-profits, government bodies, and community organizations present and utilize this data.

Action Lab, in collaboration with well-established not-for-profits, government bodies, and community organizations, is at the forefront of addressing these challenges. Our project aims to revolutionize the traditional reporting process by integrating rich multimedia content, thereby enhancing stakeholder engagement and operational transparency.

About us

Action Lab is multidisciplinary team of impact-focused researchers working at the intersection of communities, technology and social innovation. Systems designed and built at Action Lab (always in collaboration with real-world organisations and academic partners) are used nationally and internationally.

How to apply

Step 1: Find out if you are eligible and able to take it in the upcoming semester.

Step 2: Please send me an email in the following format: 

• Subject: "Interest in <TITLE>"
• Body: Explain why you are interested? What is your unique background OR what is the skillset you are trying to build for your future career?
• Attachments: A copy of your unofficial WES transcript & an up-to-date CV

Become part of the Monash Immersive Analytics Lab, and explore exciting new ways to visualise, interact, and analyse all types of data with VR and AR! We are looking for enthusiastic students to work on immersive visualisation using latest technology, such as head-mounted displays with integrated eye-trackers (Microsoft HoloLens and others), gesture recognition devices, and large wall displays. We are leaders in building internationally renowned toolkits for creating immersive data visualisations, we work with various industry partners, and we publish at top visualisation and human-computer interaction conferences. The Embodied Visualisation Group has a vibrant, collaborative research culture including fortnightly group meetings, informal and formal seminars and social events. We have a variety of thesis topics available for you to invent and evaluate new immersive visualisation and interaction techniques for large graphs, geospatial flows, building information models, maps and globes, and many other applications. Become part of our group and contact any of the members to learn more about current opportunities: https://www.monash.edu/it/hcc/embodied-visualisation/projects/active

This project focuses on modeling heart rate data captured via FitBit Sense devices worn by team members in collaborative situations such as supervision meetings, group teaching, or nursing simulation scenarios. The primary goal is to identify stressful situations or similar events by analysing heart rate variations.

This project offers a rich opportunity for IT, CS, AI, or software engineering students to delve into the practical applications of data modeling, wearable technology, and user interface design. The insights gained from this exploration can significantly enhance our understanding of stress dynamics in collaborative environments, providing valuable feedback for improving team performance and well-being.

As a participant in this project, you will be part of the Centre for Learning Analytics at Monash (CoLAM), the largest learning analytics group in the world. Our center focuses on applying AI in education and developing advanced analytics to support students and teachers. By joining our team, you will have the opportunity to work alongside leading researchers and contribute to cutting-edge projects that aim to transform educational practices through data-driven insights. This environment provides a unique platform for you to develop your skills and make a meaningful impact in the field of learning analytics and educational technology.

This project focuses on the automated classification of teachers' activities and co-teaching behaviors using positioning data captured via sensors and microphone data. The main task involves developing and applying machine learning techniques to analyse multimodal datasets, combining positioning and speech data to identify and categorize various teaching activities. By leveraging large language models (LLMs), Generative AI (GenAI), and Natural Language Processing (NLP), the project aims to extract features that enhance the accuracy and effectiveness of these classification tasks. This innovative approach will enable a deeper understanding of classroom dynamics and provide valuable insights for improving teaching practices through automated analysis and classification.

I am seeking students doing Honours or a minor thesis in a Masters interested in working on designing Learning Analytics innovations to study classroom proxemics by analysing and visualising indoor positioning data (along with other sources of data such as audio, physiological activity and characteristics of the students).

This project aims to develop methods for supporting teachers in reflecting on their positioning strategies in the classroom by making key activity traces visible. This project is fundamentally about bridging the gap between substantial work on classroom proxemics (the study about how people use the physical space), based on qualitative observations; and the dearth of methods to provide feedback to teachers on their teaching practice using evidence, at a scale. This project is strategic because it aims to transform ephemeral teaching classroom activity, that currently is largely opaque to computational analysis, into a transparent phenomenon from which selected features can be captured and rendered visible for the purposes of professional development for teachers. 

The project may involve both the analysis of a dataset already captured or the capture of new positioning data or qualitative data from educational stakeholders.

Example publication related to this opportunity: Martinez-Maldonado, R. (2019) “I Spent More Time with that Team”: Making Spatial Pedagogy Visible Using Positioning Sensors. International Conference on Learning Analytics and Knowledge, LAK 2019.

Are you ready to dive into the future of education and revolutionise how software projects are assessed? Join this innovative project aimed at creating cutting-edge learning analytics capabilities within the Faculty of IT at Monash University. This project seeks to provide automated support for teaching staff in augmenting the marking of software development and design assignments, specifically software projects submitted to the FIT-based GIT lab platform, using advanced large language models (LLMs).

By leveraging the power of GenAI, this project will transform the feedback process by analysing code revisions over time, accessing the progression of student submissions in GIT, and augmenting the qualitative analysis of submitted code for teachers and students as part of the feedback process.We will identify key teaching needs in assessment and feedback, and apply Learning Analytics and AI techniques to generate comprehensive summaries of student work. These summaries will highlight the adherence to essential design principles such as SOLID principles and design patterns, ensuring that students' code not only works but is also well-structured and maintainable.

To get a better understanding of SOLID principles and for prompt examples, check out SOLID Programming Principles.

The outcome of this project will be the development of an AI-powered teaching and feedback tool that will significantly enhance the Bachelor programs in IT, Computer Science, and Software Engineering. By streamlining the marking and feedback provision processes, we aim to improve teaching efficiency, assessment consistency, and ultimately elevate the course viability and reputation of our programs.

This project aims to enhance student engagement and comprehension by combining Data Comics with Generative AI. Data Comics present complex information in an engaging, accessible format, and by leveraging AI, we seek to automate their creation, making the process efficient and scalable. This project involves a human-centred design approach with students and teachers to ensure the content is relevant and pedagogically sound. The collaboration will tailor Data Comics to meet the needs of learners, while AI will enable the rapid generation of personalized educational materials.

Additionally, the project can explore ethical implications such as data privacy, bias, and the authenticity of AI-generated content, potentially developing a framework to address these concerns. By investigating design opportunities, including real-time feedback mechanisms and personalised learning experiences, the project aims to bridge the gap between innovative educational tools and practical classroom applications, ensuring the integration of AI and Data Comics is both effective and ethically sound.

As the number of mental health patients increases, the demand for qualified counselors is on the rise. However, training/practice sessions with actual patients are often limited, let alone meeting a sufficient number of patients of different personalities. This project aims to use large language models to simulate therapy sessions under certain predefined circumstances. This project is co-supervised by a collaborator from the Psychology department in Jeffrey Cheah School of Medicine and Health Sciences.

A recommender system is a subclass of information filtering/retrieval system that provides suggestions for items that are most pertinent to a particular user without an explicit query. Recommender systems have become particularly useful in this information overload era and have played an essential role in many industries including Medical/Health, E-Commerce, Retail, Media, Banking, Telecom and Utilities (e.g., Amazon, Netflix, Spotify, Linkedin etc).

In the past years, we have witnessed the explosive growth of generative AI tools including ChatGPT, Claude etc. This project focuses on the exploration of how generative AI can be used to boost the performance of recommender systems. Some research has been done in this area but it is still in its very early stage. 

Chatbots for mental health are shown to be helpful for preventing mental health issues and improving the wellbeing of individuals, and to ease the burden on health, community and school systems.  However, the current chatbots in this area cannot interact naturally with humans and the types of interactions are limited to short text, predefined buttons etc. In contrast, psychologists in real-world interact with patients with multiple modalities, including accustic and visual information. Non-textual information is also essential for health observation and treatments of patients.

In this project, you will contribute to data collection and implementation of the chatbot for mental health in a multi-disciplinary team. Your task will focus on the multimodal perspective of the chatbot by allowing visual and accusitic responses apart from text messages. 

Machine learning (ML) training and evaluation usually involve large-scale datasets and complicated computation. To process data efficiently, a promising solution is to outsource the processes to cloud platforms. However, traditional approaches of collecting users' data at cloud platforms are vulnerable to data breaches. Specifically, during the ML model training or inference service offering, the cloud server could learn the input data used to train the model, model structures, user queries and inference results, which may be sensitive to users or companies. 

Much research has been conducted to protect the privacy of outsourced ML tasks by employing cryptographic primitives, such as Secure Multi-Party Computation (SMC), Homomorphic Encryption (HE).  Nevertheless, SMC-based approaches do not scale well due to the large communication overhead. The efficiency of HE-based workloads is remarkably low, particularly when calculating non-linear functions. 

This project is based on the paper "Academic Journals, Incentives, and the Quality of Peer Review: A Model", in which we analyse strategic interactions between scientists and science journals.  Our results shed light on how different objectives for journals shape the strategies that scientists adopt when aiming to publish their work. In this project, we aim to extend this model to include the influence of different environmental factors such as prestige, affiliations or career stage of the scientists. We will use game theory, computer simulations and data science for validating and testing the models when possible. 

The tennis tour is a series of tennis tournaments played globally over a calendar year, where professional tennis players compete for prize money and ranking points. The structure of the tennis tour is organised into different tiers for both men and women, including grand slam tournaments and ATP/WTA tour events. In this project we use stochastic processes to model and simulate the tour under different experimental rules.  The goal is to understand how changes in the rules and structure of matches (e.g., best-of-five sets, with each set being the first to four games instead of the traditional six games) can have an impact on the way ranking points are distributed at the end of the season. 

In order to do this we will use different techniques including:
- Markov Chains and other Stochastic Processes
- Game Theory
- Computer Simulation

Description:

Magnetic Resonance Imaging (MRI) stands as a cornerstone in medical imaging, providing non-invasive, high-resolution images of the human body's internal structures.  Brain tumor segmentation from MRI scans is essential for precise diagnosis and treatment planning. MRI provides detailed views of brain structures and abnormalities, but challenges like image noise, contrast imperfections and tumor variations can make segmentation difficult.

This project focusses on using advanced deep learning techniques to accurately identify tumor regions in MRI images. By leveraging convolutional neural networks (CNNs) and other machine learning algorithms such as Vision Transformers, students are required to develop a reliable framework that automates the segmentation process, improving efficiency [1,2]. 

The objective is to streamline diagnostic workflows and facilitate timely treatment decisions by reducing manual annotation process. Through collaboration between medical imaging experts, machine learning specialists, and clinicians, this project aims to advance brain tumor diagnosis and treatment, ultimately enhancing healthcare delivery.

Students involved in this project will employ advanced image processing and programming techniques, utilize cutting edge machine learning models, and gain hands-on experience with popular tools and libraries like Python, PyTorch, and TensorFlow.

What will students learn in this project:

Participating in this deep learning assisted brain tumor segmentation project will offer students the opportunity to:

• Understand MRI Image Analysis: Gain insight into the analysis of Magnetic Resonance Imaging (MRI) scans, with a focus on identifying and delineating brain tumor regions.
• Master Deep Learning Algorithms: Learn and implement advanced deep learning algorithms, particularly convolutional neural networks (CNNs), tailored for accurate segmentation of brain tumors from MRI images.
• Practical Application of Medical Imaging: Acquire hands-on experience in applying deep learning methodologies to real-world medical imaging data, understanding their role in enhancing diagnostic accuracy.
• Programming Proficiency: Develop proficiency in programming languages and frameworks such as Python, PyTorch, and TensorFlow, essential for implementing and optimizing deep learning models for brain tumor segmentation tasks.
• Collaborative Research: Collaborate with experts in medical imaging and machine learning, contributing to cutting-edge research aimed at improving brain tumor diagnosis and treatment planning through automated segmentation techniques.
• Research & Development Skills: Develop critical research abilities, learning how to design, execute, and analyze an experimental study.
• Teamwork and Collaboration: Collaborate with a team of like-minded individuals, learning to communicate ideas effectively and working jointly towards a common goal.
• Ethical Standards in Research: Recognize the importance of maintaining high ethical standards, particularly concerning patient data and privacy.

Description:

The early detection of neurological abnormalities through Magnetic Resonance Imaging (MRI) is crucial in the medical field, potentially leading to timely interventions and better patient outcomes. However, the traditional diagnostic process is often time-consuming and subject to human error. This project seeks to improve this aspect by employing deep learning for anomaly detection in MRI scans, exclusively using images from healthy participants for model training [1].

The project aims to develop the algorithm to understand the parameters of a 'healthy' MRI scan [2], subsequently identifying deviations indicative of potential anomalies or pathological changes. This strategy significantly minimizes the bias introduced by various disease-specific patterns, enabling the model to identify unknown or unexpected anomalies efficiently.

The goal is to develop a robust, sensitive, and unbiased anomaly detection system that can assist healthcare professionals in early diagnosis and treatment planning, thereby improving patient care and management.

What will students learn in this project:

Participants in this project will submerge themselves in various learning opportunities:

• Deep Learning Techniques: Understand the nuances of deep learning algorithms and their application in anomaly detection, specifically within the realm of medical imaging.
• MRI Imaging Insights: Gain deep insights into MRI technology, focusing on the characteristics of scans from healthy individuals, understanding the subtle differences that indicate potential issues.
• Data Management and Ethics: Learn the principles of managing sensitive health data, ensuring the highest standards of privacy, confidentiality, and ethical research practices.
• Algorithm Training: Acquire hands-on experience in training deep learning models, understanding the importance of data, and nuances of 'healthy' baselines in training phases.
• Validation and Testing: Engage in the validation and testing of trained models, using unseen data to assess algorithm performance, and understand the practical challenges of real-world application.
• Research Documentation: Develop skills in documenting research findings, preparing reports, and contributing to scientific literature.
• Interdisciplinary Collaboration: Work in an interdisciplinary environment, collaborating with professionals in healthcare, machine learning, and data science.

Mental health is an ongoing issue in Australia. The cause of mental health can be due to a variety of reasons: workplace culture, high workloads, job insecurity, disparity in pay, lack of career advancement opportunities and turnover intentions. Mental healthcare workers are not able to cope with it and are suffering from burnout (Scalan et al., 2020). Most hospitals will seek agency nurses to mitigate resource constraints, but the skill sets required for the jobs are questionable. With the current COVID-19 pandemic, mental healthcare workers reported significant high levels of anxiety, depression and professional burnout (Northwood et al., 2021). Mental healthcare workers have to multitask between documentation work and attending to patients. This leads to inaccurate patient triage and missing important details during shift turnover leading to incorrect care being provided. Therefore, there is a need to ease mental healthcare workers workload and provide consistent patient triage with the help of technologies.

Problem Statement: The forensic identification of human remains is a critical legal process, culminating in the issuance of a death certificate by the appropriate authority. It is a multifaceted procedure that integrates scientific evidence—from antemortem records to advanced DNA analysis, forensic odontology, and anthropology—to match unidentified remains with missing persons. This holistic approach is foundational to resolving cases of unlawful killings, which bear significant implications for public health, legal resolution of civil affairs, and community well-being. Against this backdrop, AI with a forensic focus can potentially address the legal and humanitarian needs that arise during post-conflict periods and aiding the healing process of affected communities.

1. Identify the current state of IP violation/systematic review in popular LLM APIs such as ChatGPT and Gemini.
2. Input data IP violation: Traceability, ie.., referring to the ability to trace or track the origin and usage of data throughout its lifecycle. It involves maintaining a detailed record of where the data came from, how it has been used, and any transformations or modifications it has undergone.
3. Model IP violation: Watermarking algorithms (i.e., used to protect the models against imitation attacks).
4. Multi-modal: Recent trends have put the focus on multimodal models, which can understand the meaning of texts, images, audios and videos together in context. One example is image generation based on text description. So far, IP violations in multi-modal models have not been studied (both data and model).

This project will apply feature selection techniques for identifying features that can effectively predict the Logistics Performance Index (LPI), building upon our previously published work [1].

This project aims to expand upon the earlier research by incorporating additional distinctive features, including the carbon emissions rate, fuel and renewable energy costs and consumption rates, e-commerce market size, and growth. These elements are anticipated to provide insights into logistics performance within the contemporary landscape of a global supply chain. Furthermore, the enrichment efforts extend to the utilization of the wrapper technique.

[1]  https://doi.org/10.1007/s00521-022-07266-6

This project aims to analyse the comments of Twitter  on non-communicable diseases.  Students are expected to carry out Aspects Detection to identify the specific aspects discussed in the tweets e.g., causes, transmission and symptoms. Subsequently,  students are expected to conduct sentiment analysis utilizing tools like TextBlob or VADER, while also taking into account the importance of considering emojis to enhance classification accuracy. Students also need to provide solutions to the problem of ironic  tweets (e.g., seem positive but actually negative) and misinformation.

SCIPPlan is a mathematical optimisation based automated planner for domains with i) mixed (i.e., real and/or discrete valued) state and action spaces, ii) nonlinear state transitions that are functions of time, and iii) general reward functions. SCIPPlan iteratively i) finds violated constraints (i.e., zero-crossings) by simulating the state transitions, and ii) adds the violated constraints back to its underlying optimization model, until a valid plan is found. The purpose of this project is to improve the performance of SCIPPlan. Potential applications of this project include pandemic planning, navigation (e.g., see Figure 1 below), Heating, Ventilation and Air Conditioning control etc.

Cybersecurity researchers are contemplating how to best use the currently trending AI techniques to aid cybersecurity, beyond just for classification. 

The aim of this Honours project is to get the student to work with the supervisors on the latest AI techniques to adapt them over for cybersecurity, building first on baseline approaches for which code is available.

The student is free to discuss with the supervisor on any specific aspects of his/her choice and interest.

Cryptographic applications require a careful implementation to avoid side-channel attacks that reveal secret information to an attacker (e.g. via run-time measurements). In particular, for floating point arithmetic it is known that the timing of some basic arithmetic operations and functions on some CPUs depends on the input values [1], and thus the timing may leak secret information when the input contains secret values. Constant-time implementation tries to mitigate such run-time timing leakage on typical devices.

Recently, some newer cryptographic applications such as the Latte quantum-safe hierarchical identity-based encryption (HIBE) [2] require higher precision real number arithmetic than the double precision i.e. "double"-type variable in C. However, existing constant-time implementation techniques for real number arithmetic usually focused on the lower precision (typically within the double precision) and/or targetted one specific precision for the target cryptographic application [3,4]. Therefore, it requires extensive work to make high precision real number arithmetic constant-time and adapt the constant-time arithmetic to other cryptographic applications with different precision requirements.

Following the success of the Human Genome Project, the entire scientific community witnessed a large data explosion in genomics, which was also aided by advances in molecular biology technologies such as next-generation sequencing. These high-throughput technologies enable comprehensive molecular profiling of cancer cell lines, including gene expression. Regardless of the use of gene-based assays, they provide abundant genomic information for identifying participating genes (biomarkers) that contribute to the chemoresistance process in cancer cells.

Predictive modelling using machine learning (ML) techniques allows the generation of predictive biomarkers for chemotherapeutic response and targets for new chemotherapeutics agents. However, traditional ML techniques have some computational challenges, including (1) curse of dimensionality, (2) class imbalance, (3) heterogeneity, (4) lack of context, and (5) robust and interpretable outcomes. Therefore, a better strategy could be proposed to complement ML techniques by combining heterogeneous and large-scale datasets based on a pan-genome approach.

The need: Early detection and diagnosis of eye conditions is critically important as many diseases, including diabetic retinopathy, glaucoma, and age-related macular degeneration, often show minimal or even no symptoms. Glaucoma is called the "silent thief of sight" since it progressively damages the eyes without any noticeable signs.

This project relates to the visualisation of the source of data used in scientific experiments, and their results. The visualisation focus is graphs.

Trust in the results of scientific experiments and scientific modelling relies on knowing how they have been derived – that is, the ‘scientific workflow’ that led to their production. Being able to reproduce the scientific workflow that led to such results is critical in ensuring trust, confidence and transparency [2].

Capturing the provenance (that is, the source) of the information used for scientific workflows is therefore foundational in achieving transparency and reproducibility. There has been an increased adoption of tools such as electronic notebooks, Laboratory Information Management Systems (LIMS) and Jupyter notebooks for computational modelling that allow automated capture of provenance records. However, solutions and approaches which collate provenance information across systems in a scalable and general way are lacking.

Knowledge graph technology tools can capture concepts and the relationships between them [5, 6], and provenance ontologies are well established in the community [7]. While some prior architectures exist for implementing a knowledge graph for scientific workflows, robust implementations are not yet widespread in scientific practice [1].

This project aims to build on existing technologies and ontologies to explore how knowledge graphs can be used to represent scientific workflows, using provenance information from a variety of sources – within the context of real-world science projects. In particular, the Provena open source system curates provenance data [8], and will form the basis for the project.

Specifically, this project will explore several aspects of provenance knowledge graphs. Possible approaches include

• Developing adapters and client libraries to allow easy recording of provenance of scientific workflows, e.g. electronic notebooks, Laboratory Information Management System (LIMS), Jupyter notebooks for computational modelling.
• Developing relevant and effective queries and visualisations for knowledge discovery of scientific workflows
• Testing and benchmarking the performance of different knowledge graph implementations, e.g. neo4j [5], GraphDB [6]

An exciting prospect will be the application of this implementation to several CSIRO projects in different impact areas including Modelling national bushfire risk and resilience and Modelling interventions on the Great Barrier Reef and Hydrological modelling.

In our current day and age, there is an exponential growth in multimodal data, especially the transition of social media from text-based communications to video formats which can be observed with the rise of TikTok, Youtube, and Instagram Reels. This shift requires a shift in how we analyze multimodal data as we will have to move away from traditional text sentiment analysis such as TextCNN. Multimodal data presents us the opportunity to improve on text-based analysis given the new information that is coded in speech and visuals that can provide additional context for a sentiment.

This project has its real-world applications across different industries. Social media is one aspect which multimodal sentiment analysis can be applied to. VADER (Valence Aware Dictionary for sEntiment Reasoning) for sentiment analysis and JAMMIN for emotion analysis successfully detect hate speech on Facebook. Sentiment analysis in product reviews can greatly benefit companies trying to understand their customers sentiment for their products. In an era of generative AI hype, AI customer service models built on generative AI could benefit from understanding sentiment from a multimodal input to provide the appropriate responses.

The research challenge for this project is to curate a dataset captured in a collaborative learning setting in which teams of three students engaged in conversations and created a joint concept map. The goal is to analyse the content of their conversations and concept maps they created at a multi-touch tabletop and model the epistemic constructs reflected in both their conversations and the artefact they jointly create. Depending on the trajectory that you take, examples of the questions that such a project could investigate include:

• How can learning analytics approaches be applied to automatically model the content of conversations and the meanings embedded in students' artefacts?
• How can epistemic network analysis be applied to model the content of concept maps?
• How can epistemic network analysis be used to characterise the knowledge appropriation in high and low achieving groups? 
• How can active students and leaders provoke an epistemic shift on less active students? 

AI has been growingly used in Medicine. There are big opportunities for AI in medical research, including medical imaging diagnosis. AI and Deep Learning have been used to detect and classify lesions in various diseases, such as cancers. 

Multimedia content such as audio, image, and video are stored and transported in compressed forms. Various standards are designed to encode the content at the highest possible level while minimizing distortion. Some commonly used compression standards include MP3 for audio, JPEG for still image, H.264/AVC for video. Despite the vast differences in signal characteristics, most compression standards have two things in common: transformed-quantized coefficients and scale factor (quantization table in JPEG and AVC). The coefficients are usually coded as a product of sign_bit and magnitude. However, the sign_bit information takes up about 10% of the total file size, which is a significant portion. 

This project aims to precisely predict the sign_bit information in compressed contents - why encoded the information when we can predict it correctly? The honours student can either work on MP3 for audio, JPEG for still image, or H.264/AVC (motion vector or coefficient). Alternatively, the student is also welcome to explore the latest coding standards such as JPEG-XL for image, HE-AAC v2 for audio, and VVC for video. 

The student may choose to use handcrafted techniques (not very efficient) or deep-learning approaches. 

Podcasts have become a very popular way for small communities to create content that is meaningful for them and reach a wider audience. However, many of the skills and equipment needed to produce a good podcast are inaccessible to non-professionals and there is often a learning curve attached to gain necessary skills. In addition, the production process is seen as an individual effort (one or two producers working in isolation to produce the final edit).

This project is about designing a collaborative podcasting model with a local community (or organisation) to use in creating a podcast model for them (this includes designing a podcast production workflow in addition to training guides and helping evaluate impact of podcast). Emphasis will be put on designing a process that uses existing technologies and free-to-use tools (i.e. all recording done on smartphones and using free audio editing/mixing software).

The methodology will be inspired by Action Research, co-design and participatory design approaches, which involve a commitment to working closely with practitioners (staff and volunteers from NGOs and community organizations).

Radio is one of the primary modes in which communities across the world receive important information and build connection with wider society. Non Governmental Organisations (NGOs) have long been leveraging radio, and in particular Community Radio

In many parts of the world, audio is the preferred interface for social interactions. There has been a huge push towards audio-based interfaces for engaging marginalised communities in rural and Developing countries contexts.

This project sits within the Human Computer Interaction area, and will look at designing either an online (radio) or offline (community radio) audio-interface for connecting communities.

This can be done by using low-cost digital technologies for designing radio infrastructure.For example, devices that can be configured for this purpose include Android phones or Raspberry Pis (to ensure that the solution is low-cost and accessible).

As part of this project, you will work closely with a community organisation or NGO (this can either be an organisation that you have existing links with or we will connect you with one of our partner NGOs). Working in collaboration with the org, you will find out challenges they face in giving voice to their communities/beneficiaries that can be addressed through social media (for instance, perhaps they want to run an awareness raising campaign about the difficulties faced by the community and they want the communities to be very involved in this). Once you identify a challenge, you will work closely with them to design a process built on social media to help the organisation empower their community.

The methodology will be inspired by Action Research, co-design and participatory design approaches, which involve a commitment to working closely with practitioners (staff and volunteers from NGOs and community organizations).

See a previous example of this approach: WhatFutures <https://whatfutures.org>

This project is within the scope of the project “Artificial Intelligence in carDiac arrEst” (AIDE), which was led by Ambulance Victoria (AV) in Australia, involving a team of researchers at Monash University. This AIDE project has developed an Artificial Intelligence (AI) tool to recognise potential Out-of-Hospital-Cardiac Arrest (OHCA) during the Triple Zero (000) call by using transcripts produced by Microsoft Automatic Speech Recognition service. Within the framework, the student will extend the tool to recognise OHCA effectively by using both transcripts and audio signals, and notify the call-taker of the level of probability of a cardiac arrest at the earliest possible point of recognition.

Recently, program generation and optimisation techniques have been adapted to performance critical subroutines in cryptography. Codes generated/optimised by these techniques are both secure and their performance is highly competitive compared to hand-optimised code by experts [1].

IT Forensics is the art of extracting digital pieces of evidence also known as (aka) artifacts in a forensically sound manner, that is presentable to a court of law. In doing this it covers a range of conceptual levels, from high-level operating systems and computer theory down to computer networking. 
 

The specific objective(s) of this project is to look at an encrypted piece of data and distinguish what encryption algorithm is used/employed. This would benefit IT Forensics researchers/investigators attacking encrypted volumes, files, folders, etc.

Since the 1990s, researchers have known that commonly-used public-key cryptosystems (such as RSA and Diffie-Hellman systems) could be potentially broken using efficient algorithms running on a special type of computer based on the principles of quantum mechanics, known as a quantum computer. Due to significant recent advances in quantum computing technology, this threat may become a practical reality in the coming years. To mitigate against this threat, new `quantum-safe’ (a.k.a. `quantum-resistant’ or `post-quantum') algorithm standards for public-key cryptography are in development [1], that are believed to be resistant against quantum computing attacks.

Mini-CP https://www.info.ucl.ac.be/~pschaus/minicp.html is a minimal form of constraint programming solver, designed to allow for easy experimentation and learning. 

One of the most efficient approaches to discrete optimisation solving is using lazy clause generation, which is a hybrid SAT/CP approach to solving problems.  But MiniCP does not currently support this. 

Using a version of Mini-CP written in C++ (https://ldmbouge@bitbucket.org/ldmbouge/minicpp.git), this project will build a Mini-LCG (minimal LCG solver) by combining MiniSAT http://minisat.se/ a venerable but clean SAT solver with MiniCP in C++.  

Given a knowledge base describing the existing background constraints and assumptions about what is possible in the world as well as the prior experience of an autonomous agent on the one hand and probabilistic perception of the current state of the world of the autonomous agent, on the other hand, it is essential to devise and efficiently enumerate the most consistent world models that are likely to be valid under the prior knowledge in order to refine the agent’s up-to-date perception and take the most suitable actions. This project will exploit modern answer set programming (ASP), constrained ASP (CASP), core-guided maximum satisfiability (MaxSAT) reasoning, bounded model checking (BMC) and relate with Markov Logic Networks (MLN) in order to facilitate interaction between symbolic knowledge and a neural agent and succeed in overall neuro-symbolic tasks, i.e. those posed by Animal AI-Testbed. Students in this project will collaborate closely with those involved in the project on “Building consistent world states for an autonomous agent”.

Building a robust and trustworthy (semi-)autonomous agent requires us to build a consistent picture of the state of the world based on the data received from some perception module.

In this project we explore the use of modelling formalisms like Answer Set Programming (ASP) and Constraint ASP (such as the one implemented in the s(CASP) system)  to encode the background knowledge and commonsense reasoning required for our agent to succeed in tasks such as those posed by the Animal AI-Testbed. This will require us to  explore the use of probabilistic-aware, non-monotonic rules, facts and constraints that can be used by the agent to build a world state (i.e., a static snapshot of the environment at some time T), while also maintaining a historical view formed by previous snapshots. Each world state must contain a consistent set of objects, their information and the associated relations among these objects.

Students in this project will collaborate closely with those in projects “Efficient exploration of consistent worlds” and “Solving Automata using ASP/Minizinc?”

In this project, you will build an autonomous agent in the MineRL environment for playing Minecraft or an agent for Animal-AI.  Herein, you will learn how to incorporate symbolic prior knowledge for improving the performance of an agent trained by using deep reinforcement learning (RL) technique, which is the core technique to build AlphaGo. An RL-based agent learns a stochastic policy to decide which action to take in the next step. Correct choices of actions will be rewarded by the gaming environment. Symbolic knowledge will be used to build a constraint action space for an agent at each time step such that infeasible actions are excluded through (defeasible) reasoning. As a result, an agent would benefit from both symbolic knowledge and strengths of deep reinforcement learning.     

Project duration (between 1 & 12 weeks): 12 weeks 

Prerequisites: FIT2014, MAT1830, FIT3080, FIT5201

The existing deep learning-based time series classification (TSC) algorithms have some success in multivariate time series, their accuracy is not high when we apply them to brain EEG time series (65-70%). This is because there is a large variation between EEG data of different subjects, so a TSC model cannot generalise on unseen subjects well. In this research project, we investigate self-supervised contrastive learning to encode the EEG data. This way we can better model the distribution of our EEG data before classifying it into different mental statuses. See recent work here [1]. The data that will be used in this project includes (but is not limited to), data captured from Emotiv [2] brainwear devices. 

[1] Kostas, D., Aroca-Ouellette, S., & Rudzicz, F. (2021). BENDR: using transformers and a contrastive self-supervised learning task to learn from massive amounts of EEG data. Frontiers in human neuroscience.

[2] https://www.emotiv.com

Turning Point’s National Ambulance Surveillance System is a surveillance database comprising enriched ambulance clinical data relating to alcohol and other substance use, suicidal and self-injurious thoughts and behaviours, and mental health-related harms in the Australian population. These data are used to inform policy and intervention design and are the subject of ever-increasing demand from academic professionals and units, government departments, and non-government organisations.

Develop, implement, and test deep learning techniques for automatic classification of epileptic seizures using video data of seizures

This project takes a different approach to RL, inspired by evidence that Hippocampus replays to the frontal cortex directly. It is likely used for model building, as opposed to the mainstream view in cognitive science and ML - where 'experience replay' ultimately improves policy. The predicted benefits are sample efficiency, better ability to generalize to new tasks and an ability to learn new tasks without forgetting old ones. The project objective is to improve biological models and advance state-of-the-art in continual reinforcement learning.

The brains of all bilaterally symmetric animals on Earth are divided into left and right hemispheres. The anatomy and functionality of the hemispheres have a large degree of overlap, but they specialize to possess different attributes. This principle is poorly understood and has not been exploited in AI/ML. The right hemisphere is more dominant for novelty, and the left for routine. Activity slowly moves to the left hemisphere as a task is perfected. In this project, we apply that principle to continual RL, where new tasks are introduced over time. We will create a ‘generalist’ right network that can perform novel tasks while a left network has time to become proficient, providing a more maintained level of competence across new tasks - a critical characteristic for practical agents to operate in realistic environments.

The hippocampus is critical for episodic memory, a key component of intelligence, and a sense of self. There are a number of computational models, but none of them consider the fact that the hippocampus is, like the rest of the brain, divided into Left and Right hemispheres. Division into Left and Right is poorly understood, but undoubtedly critical, as it is a remarkably conserved feature of all bilaterally symmetric animals on Earth. Previously, on a non-episodic classification problem, we mimicked biological differences between hemispheres in left and right neural networks, and achieved specialization and superior performance that matched behavioral observations. This project asks how specialization can improve episodic or one-shot learning by creating a hippocampal model with left and right neural networks. This will be a truly novel approach to hippocampal modeling, and will help with one of the biggest mysteries in cognitive science, ‘why are brains divided into left and right?’. It also constitutes a new principle in AI/ML.

The brains of all bilaterally symmetric animals on Earth are divided into left and right hemispheres. The anatomy and functionality of the hemispheres have a large degree of overlap, but they specialize to possess different attributes. This principle is poorly understood and has not been exploited in AI/ML. Previously, we mimicked biological differences between hemispheres, and achieved specialization and superior performance in a classification task that matched behavioral observations. Similar mechanisms are likely to underpin specialization observed in motor control, where one side specializes in the control of trajectories and the other in the control of posture. This project investigates that question by building a model with left and right neural networks to perform a motor task, and compare to human performance, and standard ML approaches. This will help with one of the biggest mysteries in cognitive science, ‘why are brains divided into left and right?’, and constitutes a new principle in AI/ML.

Using relevant available data-sets, we compare appliance usage across households of different demographics.  We then use machine learning techniques to infer how different households use different appliances at different times, resulting in diverse energy consumption behaviours. 

Climate change will affect us all, and we have to do everything we can to minimize the magnitude of change. Investments in renewable generation help to reduce the impact of energy usage on the supply side, but that will not get us all the way there, especially in the near term. Consumers will also have to become much more efficient with their energy use.

To become an efficient user of electricity, consumers need to know where their electricity usage comes from. Unfortunately, our current monthly electricity bill does not accurately reflect what drives our usage. What’s needed are machine learning techniques to derive detailed usage information from aggregate electricity measurements taken by smart meters, a process known as non-intrusive load monitoring (NILM).

In this project, you will study and improve machine learning techniques - including Hidden Markov Models (HMMs), Long Short-Term Memory (LSTM) deep learning networks and/or applications of the Bayesian information-theoretic Minimum Message Length (MML) principle - to model consumer electricity usage. We intend to improve accuracy by integrating additional features of known patterns of use (such as time-of-day associated with activities) and/or characteristic load patterns (such as transient load decay associated with refrigerator cooling cycles).

Behavioural manifestations of epileptic seizures (ESs) and certain non-epileptic seizures (psychogenic non-epileptic seizures, or PNESs) have considerable overlap, and so discerning between these solely based on clinical criteria is difficult.  Video EEG (electroencephalogram) monitoring (VEM) has high resource demands and is also expensive.  We endeavour to classify seizures based on non-invasive measures.

The security threat by quantum computing to almost all currently used digital signatures was triggered by the discovery of Shor’s quantum algorithm, which efficiently breaks the two problems underlying the security of these schemes, namely integer factoring, and elliptic curve discrete logarithms (ECDLP). When quantum computers become widespread, all security for the current digital signatures that are widely used to secure a wide range of systems is lost. With quantum computing as a software service offered by tech giants, any entity using them will be able to employ these services to forge/attack any digital signature of interest. 

It is believed that the scale of cryptographically relevant quantum computing capability has a significant risk of realizability within the next 15 years. Consequently, governments around the world (such as Germany, through the German Federal Office for Information Security - BSI) are actively working on post-quantum cryptography and are promoting the adoption of post-quantum systems for high-security applications [White Paper]. In a recent publication titled "Commercial National Security Algorithm Suite and Quantum Computing FAQ", the National Security Agency (NSA) of the United States warned of the potential security threat posed by quantum computers and advocated the need to act urgently to protect national security services. Among potential candidates for quantum-safe cryptography, NSA identified lattice-based schemes [NSA] as one of the most efficient candidates.

Later, Google announced its experimental realisation of quantum-safe cryptography protocols for the Chrome web browser. In response, to ensure the security of cryptography into the future, the US National Institute of Standards and Technology (NIST) initiated a 5-10 year standardisation process [NIST] at the end of 2017 for quantum-resistant cryptographic algorithms to be evaluated for selecting the new public-key cryptography standards.
This process is now in its security evaluation stage, as the understanding of the security of cryptography against quantum computing attacks is still incomplete. 

This multidisciplinary project combines cutting-edge Natural Language Processing (NLP), Chinese Studies and Political Science. The project aims to develop a deeper understanding of how official discourse has developed throughout the history of the People’s Republic of China. The main focus will be on text in the People’s Daily, the largest newspaper in China and the official newspaper of the Chinese Communist Party. The People’s Daily plays a major role in the Chinese government’s communication with the public, and it is therefore a key resource for understanding the process of governance.

The project draws on insights from previous qualitative Social Science research in Chinese Studies and Political Science, by researchers who understand Chinese context. Previous research in Chinese Studies has examined how key concepts with deep cultural resonance form part of the government’s discourse. For instance, the concept of suzhi (素质), which is approximately translated as ‘quality’ citizenship, has been used repeatedly throughout the recent history of Chinese governance in different contexts and policy areas, and with subtly different and changing meanings. Qualitative research has revealed insights into these changing meanings. NLP methods that build on these insights have the potential to reveal even more patterns and trends.

The student will be part of a supportive multidisciplinary team that includes established and emerging researchers in the fields of NLP, Chinese Studies and Political Science. The main supervisors will be Dr. Yuan-Fang Li (Data Science, Monash) and Professor Robert Thomson (Political Science, Monash). Other members of the team will be Associate Professor Delia Lin (Chinese Studies, University of Melbourne), Ms. Yang Wang (Chinese Studies, University of Melbourne), and Mr. Xinwei Chen (Political Science, Monash).

The student’s project will support the work of this multidisciplinary research team, by contributing to their ongoing work that addresses key research questions on official discourse on China.

Commonsense reasoning refers to the ability of capitalising on commonly used knowledge by most people, and making decisions accordingly. This process usually involves combining multiple commonsense facts and beliefs to draw a conclusion or judgement. While human trivially performs such reasoning, current Artificial Intelligence models fail, mostly due to challenges of acquiring relevant knowledge and forming logical connections between them. This project aims to develop and evaluate machine learning models for commonsense reasoning, with question answering as the key application. 

Developing quality AI tools for legal texts is the focus of enormous industry, government and
scholarly attention. The potential benefits include greater efficiency, transparency and access to justice.
Moving beyond the hype requires novel transdisciplinary effort to combine IT and Law expertise.
This project engages this challenge by developing a semi-structured knowledge base (KB) and
reasoners for statutes and cases. The project will also construct corresponding training and evaluation
datasets.

This project is within the scope of the project “Artificial Intelligence in carDiac arrEst” (AIDE), which was led by Ambulance Victoria (AV) in Australia, involving a team of researchers at Monash University. This AIDE project has developed an Artificial Intelligence (AI) tool to recognise potential Out-of-Hospital-Cardiac Arrest (OHCA) during the Triple Zero (000) call by using transcripts produced by Microsoft Automatic Speech Recognition service. In the next step, we aim to optimise the workflow of call-takers and investigate which workflows can lead to earlier identification of OHCA. This is still an open challenge of call centre triage and requires counterfactual analysis of call histories. In this project, the student will develop the counterfactual analysis tool to find out if OHCA can be identified earlier if call-takers had asked different questions. 

A lot of decision support systems have been developed to predict or suggest a diagnosis about the health conditions of patients with the aim to assist clinicians in their decisional process. One of the techniques that is proved to present an efficient tool for medical healthcare decision making is Bayesian networks (BNs). BNs are recognized as efficient graphical models that can be used to explain the relationships between variables. BNs have significant capabilities for investigating biomedical data either to obtain relationships between biomedical risk factors or either for medical predictions. Within healthcare domain, there has been some cross disciplinary research that have attempted to employ BNs technique in predicting the presence of mood and psychosocial disorders. There are several studies, for instance, that have examined the causal relationships between symptoms and depression.

Digital signatures are asymmetric cryptographic schemes used to validate the authenticity and integrity of digital messages or documents. The signer uses their private key to generate a signature on a message. Then, this signature can be validated by any verifier who knows the signer’s corresponding public key. Sometimes a digital message might require signatures from a group of signers. The naïve method to achieve this goal is collecting distinct signatures from all signers. Multisignature schemes enable a group of signers to jointly generate a common signature that is more compact than the output from the mentioned naïve method. For some multi-signature schemes, verifiers are able to aggregate the signers’ public keys into one aggregated public key, as well. Then, signature verification can be done like a regular signature verification. This property is called key aggregation and is useful for use cases where public keys size matters (e.g. Blockchain and/or cryptocurrency). Multisignature might also be used in IoT applications with resource-constrained IoT devices. To design a multisignature scheme for such use cases, storage, memory, processing, and battery lifetime requirements of IoT devices must be considered.

Visually discriminating the identity of multiple (similar looking) objects in a scene and creating individual tracks of their movements over time, namely multi-object tracking (MOT), is one of the basic yet most crucial vision tasks, imperative to tackle many real-world problems in surveillance, robotics/autonomous driving, health and biology. While being a classical AI problem, it is still very challenging to design a reliable multi-object tracking (MOT) system capable of tracking an unknown and time-varying number of objects moving through unconstrained environments, directly from spurious and ambiguous measurements and in presence of many other complexities such as occlusion, detection failure and data (measurement-to-objects) association uncertainty.

The ability to forecast human trajectory and/or body motion (i.e. pose dynamics and trajectory) from camera or other visual sensors is an essential component for many real-world applications, including robotics, healthcare, detection of perilous behavioural patterns in surveillance systems. However, this problem is very challenging; because there could potentially exist several valid possibilities for a future human body motion in many similar situations and human motion is naturally influenced by the context and the component of the scene/ environment and the other people's behaviour and activities. In this project, we aim to develop such a physically and socially plausible framework for this problem.

Human behaviour understanding in videos is a crucial task in autonomous driving cars, robot navigation and surveillance systems. In a real scene comprising of several actors, each human is performing one or more individual actions. Moreover, they generally form several social groups with potentially different social connections, e.g. contribution toward a common activity or goal. In this project, we tackle the problem of simultaneously grouping people by their social interactions, predicting their individual actions and the social activity of each social group, which we call the social task. Our goal is to propose a holistic approach that considers the multi-task nature of the problem, where these tasks are not independent and can benefit each other.

3D localisation, reconstruction and mapping of the objects and human body in dynamic environments are important steps towards high-level 3D scene understanding, which has many applications in autonomous driving, robotics interaction and navigation. This project focuses on creating the scene representation in 3D which gives a complete scene understanding i.e pose, shape and size of different scene elements (humans and objects) and their spatio-temporal relationship.

In this project, the goal is to develop a new method (using computer vision and machine learning techniques) for robotic navigation in which goals can be specified at a much higher level of abstraction than has previously been possible. This will be achieved using deep learning to make informed predictions about a scene layout and navigating as an active observer in which the predictions inform actions. The outcome will be robotic agents capable of effective and efficient navigation and operation in previously unseen environments, and the ability to control such agents with more human-like instructions. Such capabilities are desirable, and in some cases essential, for autonomous robots in a variety of important application areas including automated warehousing and high-level control of autonomous vehicles.

To operate, interact and navigate safely in dynamic human environments, an autonomous agent, e.g. a mobile social robot, must be equipped with a reliable perception system, which is not only able to understand the static environment around it, but also perceive and predict intricate human behaviours in this environment while considering their physical and social decorum and interactions.

The Optimisation group is looking for multiple students to contribute to our world leading research. Our interests range from practical to theoretical. So whether you are interested in path finding for AI in games, solving a complex scheduling problem, designing new algorithms, or working on our specialised modelling language, we will have a project that is of interest to you!

Examples of projects that have been completed by summer students within the optimisation group include:

• MiniZinc testing infrastructure: The optimisation group designs, develops and maintains MiniZinc, a free and open-source constraint modeling language with thousands of users around the world. As part of a summer project, a student has designed a testing infrastructure that helps avoid the occurrence of bugs in the compiler.
• Woodside pipe layout scheduling: To create the most efficient chemical production plant, Woodside collaborates with Monash to explore optimisation techniques capable of optimising the design of these plants. Summer students have helped to improve and visualize the proposed solutions generated by these techniques.
• Pathfinding for games: In computer games we use AI algorithms to help characters navigate their virtual environments. Although simple in principle these problems are surprisingly tricky in practice: there can be hundreds of moving agents, the environments are often large and constantly changing, and most of the available compute resources (CPU, memory) are prioritised elsewhere -- such as toward graphics! Students working on this project have developed new and influential pathfinding techniques, including ones that are used in current AAA games.  

Despite the rapid progress made recently, deep learning (DL) approaches are data-hungry. To achieve their optimum performance, a significantly large amount of labeled data is required. Very often, unlabelled data is abundant but acquiring their labels is costly and difficult. Many domains require a specialist to annotate the data samples, for instance, the medical domain. Data dependency has become one of the limiting factors to applying deep learning in many real-world scenarios. As reported, it costs more than 49,000 workers from 167 countries for about 9 years to label the data in ImageNet, one of the largest visual recognition datasets containing millions of images in more than 20,000 categories. To make the training and evaluation process of DL applications more efficient, there is an increasing need to make the most out of limited available resources and select the most valuable inputs for manual annotation. This project aims to addresses the problem of prioritizing error-revealing samples from a large set of unlabelled data for various DL tasks.

In multi-label classifications, a data sample is associated with more than one active label, which is a more challenging task than conventional single-label classifications. This project will focus on eXtreme Multi-Label (XML) classifications for text data (i.e., documents), where the label set can be extremely large, e.g., more than 10,000. For example, the input texts can be the item descriptions of an e-commerce website (e.g., Amazon) and one needs to classify them into a large set of item categories. The project is to develop novel machine learning and deep learning models for XML of text data by leveraging metadata of documents and knowledge in pretrained language models.

Over the past decades, we have witnessed the emergence and rapid development of deep learning. DL has been successfully deployed in many real-life applications, including face recognition, automatic speech recognition, and autonomous driving, etc. However, due to the intrinsic vulnerability and the lack of rigorous verification, DL systems suffer from quality and security issues, such as the Alexa/Siri manipulation and the autonomous car accidents, which are introduced from both the development and deployment stages. Traditionally, testing and verification are applied to improve the quality of software systems, either to find defects or to prove they are bug-free. However, due to the fundamentally different programming paradigm and logic representation from traditional software, existing quality assurance techniques can hardly be directly applied to DL systems. In this project, you are able to learn about the state-of-the-art analysis methods of deep learning and investigate how they perform on different types of neural networks.

The research challenge for this project is to research, prototype and evaluate approaches to automatically capture multimodal traces of team members’ activity using sensors (such as indoor positioning trackers, physiological wristbands and microphones), using learning analytics techniques to make sense of sensor data from healthcare contexts. Depending on the trajectory that you take, examples of the questions that such a project could investigate include:

• How can multimodal analytics approaches be applied to gain a holistic understanding of team members’ activity in authentic learning / training spaces?
• How can the insights of team members’ activity in physical (e.g. classrooms) or digital (e.g. Zoom) spaces be connected with higher-level team constructs?
• How can these insights promote productive behavioural change?
• How can the facilitator be supported with this information to provide informed feedback?
• What are the ethical implications of rolling out teamwork analytics in the classroom or team settings?
• How can this information enable the assessment of teamwork?
• How do learning theories or teamwork theory can inform the design of such analytics tools?

The following paper can serve as an illustrative example of this strand of research:

Towards Collaboration Translucence: Giving Meaning to Multimodal Group Data. CHI 2019 [PDF]

The aim for this project is to research, prototype and/or evaluate approaches to increase the explanatory effectiveness of the visualisations contained in analytics dashboards or similar support data-intensive tools. Explanatory visualisations are those whose main goal is the presentation and communication of insights. By contrast, exploratory visualisations are commonly targeted at experts in data analysis in search of insights from unfamiliar datasets. The premise is that most of current analytics tools are not designed as explanatory interfaces. Here we are talking about users without data analysis training who may need to interact with data (for example, using a dashboard). This is an area that can lead to important contributions in the areas of learning analytics and information visualisation. We sit at the Centre of Learning Analytics at Monash so you may want to focus on educational contexts. But there is an option to focus on a more general context (e.g. using alternative datasets). 

Disruptive technologies such as artificial Intelligence (AI) systems can have unintended negative social and business consequences if not implemented with care. Specifically, faulty or biased AI applications may harm individuals, risk compliance and governance breaches, and damage to the corporate brand. An example for the potential harm inflicted on people is the case of Robert Williams who was arrested because of a biased insufficiently trained facial recognition system in the US in 2020 (See the New York Times link below).

In this project, we aim at surveying relevant computational tools/models used for automatic question generation, and then comparing the effectiveness of these tools/models by using existing datasets.

Amphetamine (AMPH) is a widely abused drug, but before it was restricted in use it was an effective
weight loss agent. We have shown that a distinct group of neurons controls a large part of the
body’s weight loss response to amphetamine. In 2017 a single cell RNA sequencing project was
published (Campbell, Macosko et al. 2017) that described transcriptional profiles of 21,000 neurons,
amongst which are the neurons we have shown mediate the weight loss caused by amphetamine.
Amphetamine does not act directly on these neurons, rather AMPH increases the concentration of
the monoamines dopamine, noradrenaline and serotonin in the brain. We wish to analyse the
Campbell data set 1 to determine which sub populations of our neurons of interest express
monoamine neurotransmitter receptors.
Initially we anticipate using Seurat and principle component analysis clustering to tease apart the
various population of relevant neurons, and then test for the expression of genes that encode
monoamine receptors, because it is these monoamine receptors that mediate the metabolic (weight
loss) effects of AMPH. We anticipate that the identification of appropriate populations of neurons
will enable direct targeting of those neurons by chemogenetic techniques to test their effects on
metabolism. This will allow us to determine if those neurons are targets for further drug
development. We seek a Masters level bioinformatics student to help with the analysis of this data
and the design of future single cell RNA sequencing studies to help us identify new drug targets,

Campbell, J. N., E. Z. Macosko, H. Fenselau, T. H. Pers, A. Lyubetskaya, D. Tenen, M. Goldman, A. M. Verstegen, J. M. Resch, S. A. McCarroll, E. D. Rosen, B. B. Lowell and L. T. Tsai (2017). "A molecular census of arcuate hypothalamus and median eminence cell types." Nat Neurosci 20(3): 484-496.

Raw Drop-seq data and processed DGE files are available at GEO accession code GSE93374.

The evolutionary back and forth between hosts and mobile genetic elements drives the innovation of remarkable molecular strategies to sense or conceal foreign genetic material. The Knott Lab uses bioinformatics, biochemistry, and structural biology to understand how CRISPR-Cas and other novel immune systems specifically sense DNA or RNA. We aim to better understand the function of nucleic acid sensors to harness their activity as tools for molecular diagnostics or as innovative biomedicines.

The proposed project will involve mining available metagenomic data to expand on and discover novel immune systems that specifically target RNA. Students will have the opportunity to explore bacterial and phage metagenomes to compile datasets of interesting RNA sensors. Students will use these datasets to generate large multiple sequence alignments and phylogenetic trees to reveal associations between existing and novel RNA sensors. The resulting datasets will uncover potentially novel and exciting nucleic acid sensors and serve as the foundation for subsequent biochemical experiments to assay function.

Pancreatic ductal adenocarcinoma (PDAC) is one of the most aggressive forms of cancer, with a high mortality rate. Therapies for PDAC are limited to chemotherapy, which has been ineffective to treat advanced stages of the cancer. Therefore, it is critical to identify biomarkers and develop targeted therapies, to improve early diagnosis and patient care. We have generated mass spectrometry (proteomics and phosphoproteomics) data from 15 PDAC patient-derived xenographs (PDX), which have been grown in mice. These PDX can be separated based on their response to chemotherapy. The project aim is to develop a bioinformatics pipeline to identify the proteomics-based signature differentiating chemo-resistant versus sensitive tumours. Ultimately, the data will be used to identify biomarkers and potential therapeutic targets for further evaluation. The student will use bioinformatic techniques to cluster protein expression profiles, enrichment of signalling pathways and design data visualisation approaches to dissect the complex real-life biological data.

Immune protection provided by immune memory underpins successful vaccines and is mediated mainly by memory lymphocytes and long-lived antibody- secreting cells. In particular, B cell memory is key to providing a rapid and robust response upon secondary infection and continual serum antibody protection. We are working to elucidate the crucial epigenetic mechanisms that generate and maintain B cell memory, and how B cells may retain molecular and functional plasticity under chronic pathogenic pressure. In particular, chronic infectious diseases in which the pathogen is not effectively cleared from the body, such as HIV and hepatitis C, have posed longstanding challenges to global health. Thus, there is a critical need for new strategies to prevent and/or treat chronic viral infection.

We have taken a molecular approach to unlocking memory B cell potential. We will apply the cutting-edge techniques single cell RNA-sequencing and single cell ATAC-sequencing to understand the epigenetic regulation of B cell diversity. This Master of Data Science Project will analyse and integrate these single cell datasets using Seurat, then undertake sequence motif analysis and transcription factor footprinting. Pseudotime analysis will be undertaken to visualise the trajectory of the cells in acute and chronic infection, enabling us to determine what key changes occur that may lead to the reprogramming of the memory B cell population during chronic infection. We will investigate whether we can use this analysis to predict a divergence point where memory B cell changes in chronic infection become irreversible.

Antimicrobial resistance poses significant medical challenge worldwide. Misuse, overuse or suboptimal dosing of antibiotics are major driving factors of antimicrobial resistance. Pharmacokinetic/pharmacodynamic (PK/PD) modelling is critical for designing optimal antimicrobial therapies to maximise the efficacy and minimise the emergence of resistance. However, conventional PK/PD modelling is generally based on viable counting on agar plates after overnight culture and employs a population approach. Therefore, a high-throughput, real-time approach is required to capture the dynamics of antimicrobial killing and resistance at single-cell level for accurate antimicrobial PK/PD modelling.
We have developed a time-lapse microscopy method using a microfluidic chamber to record morphological changes of bacterial cells following antimicrobial treatment. However, automatic cell lineage tracking is challenging and necessitates the development of a robust and efficient computational tool to solve this problem.
This multidisciplinary project will develop and validate an advanced computational toolkit to automate cell segmentation, cytometry analysis, cell lineage tracking and single-cell based PK/PD modelling. The developed imaging-based single-cell PK/PD approach will accurately capture bacterial population heterogeneity following antimicrobial treatment, and shift the paradigm of conventional antimicrobial pharmacological research.

Antimicrobial resistance (AMR) has posed critical challenges to global health. The World Health Organization has identified carbapenem-resistant Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacterales as the top-priority pathogens urgently to be targeted for the development of novel therapeutic options. Recently, bacteriophage therapy has attracted extensive attention owing to its potential as novel antimicrobials to combat MDR pathogens.
Bacteriophages are the most abundant and diverse organisms in the biosphere. To date, there are over 10,000 complete phage genomes in the GenBank database. Despite their relatively small genomes, bacteriophages show extraordinary genomic diversity and complex evolutionary relationships that do not follow traditional hierarchical phylogeny. The limited understanding of phage genomes and evolutionary relationship has significantly hindered the development and optimisation of effective bacteriophage therapy against ‘superbugs’.
This project will conduct comparative genomic and evolutionary analyses of bacteriophages infecting top-priority Gram-negative pathogens. The available bacteriophage genomes from public databases will be combined with our own sequenced genomes for comprehensive analysis. Evolutionary analysis for core genes will be conducted and compared with the phylogeny of host bacteria to investigate the bacteriophage-bacteria interactions in the context of coevolution. Better understanding phage genomic diversity and evolutionary conservation will facilitate the development of innovative phage therapy against the problematic ‘superbugs’.

Spontaneous synchronization is a common phenomenon occurring in diverse contexts, from a group of glowing fireflies at night or chirping crickets in a field to a network of coupled neurons in the brain. The study of synchronization helps to understand how uniform behaviors emerge in populations of heterogeneous neurons. At a macroscale level, the cortex operates in two classically-defined states: “synchronized” state which is characterized by strong low-frequency fluctuations and “desynchronized” state in which low-frequency fluctuations are suppressed. These states determine how information is processed in the cortex. Similarly, at microscale level significant correlations in fluctuations of spontaneous spiking activity across sensory cortex neurons has been observed. The aim of this project is to characterize the non-stationary dynamics of activity at these two levels, and they are linked across different brain regions over space and time. We hypothesise these dynamics reflect the function of neural networks in evoked conditions.
Multi-electrode electrophysiology data from multiple brain areas is available for interested students. Additionally, simulation and computational projects to simulate the behaviour of coupled networks is available. work with animals is not required, however, if interested, there will be a unique opportunity to observe or contribute in animal experiments.

The aim of this project is to understand the computations underlying animals’ choice in dynamic and changing environments. The natural environment is multisensory, dynamic and changing, requiring animals to continually adapt and update their learned knowledge of statistical regularities in the environment that signal the presence of primary needs like water, food and mates. Yet, how the brain adapts and updates itself to the non-stationary and dynamic attributes of natural environments remains unexplored. We trained rats in a two-choice sensory categorisation task that the categorisation boundary switches between two values. choices were rewarded by diluted fruit juice. The rats followed the change in the boundary after 20-30 trials. However, the animal’s choices in a proportion of trials depends on parameters other than the current stimulus. These factors include the history of previous choices, the outcome of previous choices (correct vs incorrect categorisation of stimulus). The aim of this project is to quantify the behavioural states where the choices are governed by non-sensory components such as attentional lapses and bias, or sensory cue, and then provide a mechanistic model that explains the behaviour of the animals. Direct work with animals is not required, however, if interested, there will be a unique opportunity to observe or contribute in animal experiments.

Recent technological advances in micro and nano-fabrication technology and high-yield electrophysiology techniques allowed us to record the activity of hundreds/thousands of neurons simultaneously. This has spurred renewed interest in applying multi-electrode extracellular electrophysiology approaches in the field of neuroscience. Each electrode samples the activity of one or more neurons in its vicinity. One of the major challenges is to efficiently and robustly detect the spikes that individual neurons fire from the raw recorded electrophysiological signals. Current methods are computationally demanding, slow, unreliable in noise rejection, sensitive to optimal selection of parameters and usually require human supervision. The aim of this project is to develop a new spike sorting approach using machine learning and deep learning methods. The desire method will identify spikes, remove artefacts and noise from raw data including photoelectric artefacts from optogenetics and imaging or motion artefacts from cables and movement of animals, and predict the occurrence of spikes based on other electrophysiological measures. Direct work with animals is not required, however, if interested, there will be a unique opportunity to observe or contribute in animal experiments.

With up to 1 in 9 Australians affected and an incidence on the rise, there is a clear need to understand the mechanisms driving asthma. This research project aims to dig deep into the early origins of this disease using cutting-edge sequencing technologies in order to identify targets that could be the focus of new therapies and prevention strategies. Historically, studies have focused on one specific aspect of the disease; for example genetics and heritability, environmental factors, microbiome, or respiratory infections. But asthma is a complex multifactorial disease and each individual approach cannot capture the entire pathophysiological mechanisms underlying its development. By using the latest advances in ‘omics’ technologies on a unique sample set of asthmatic children, this project aims to provide an integrative view of the early origins of the disease. More specifically, we aim to discover molecular biomarkers of the disease (microbial, immune and metabolic) prior to its development which would serve as a platform for novel therapeutic intervention strategies that would redirect infants on a high-risk trajectory towards health.
The student will be working on establishing bioinformatics pipelines on a high performance computing cluster for dual high throughput DNA and RNA sequencing data analyses and multi-omics data integration. Coding skills in R and bash are required and completion of advanced units in bioinformatics preferred.

The DNA inside a cell is not randomly distributed but rather organized in a structure called chromatin. This non-random distribution has important implications for the functioning of cellular programs. The basic building block of this organisation system is the nucleosome. The nucleosome consists of a short piece of DNA wrapped around a protein core, with millions of nucleosomes are present in the cell’s nucleus. The orientation of nucleosomes with respect to each other and the way they pack the genomic DNA determine the architecture of chromatin. The 3D architecture of chromatin influences which genetic programs can be accessed at a given point in time and is, therefore, fundamental for all biological processes in the cell.

We are working to determine the 3D structure of chromatin and how it affects the regulation of gene expression. The student would work to establish relative orientations and distances between nth-nearest neighbours nucleosomes and search for patterns of chromatin geometry in large data sets defining the structure of chromatin in its distinct states. The ultimate goal of the project is to establish a streamlined pipeline for the objective mapping and measures of chromatin structure and geometry, according to variables defined during the project.

Neuroscience is becoming an exciting and multidisciplinary field, with a combination of biology, psychology, engineering, and large-data processing. This project is suitable for those who are motivated to apply data-processing skills to biological questions. Our research projects aim to investigate how neural circuits in the mouse brain work during a behavioral task; we visualize neural activity in vivo using advance fluorescent microscopy (two-photon imaging), while filming the behavior of mice. Animal movements need to be classified accurately and efficiently from the video, using deep learning (DeepLabCut), and to be linked to neural activity in vivo.

Chemical exchange saturation transfer (CEST) MRI provides images of molecular information and has recently been used for the detection of malignant brain tumors and the assessment of muscle tissue in cardiac infarction. Additionally, CEST has also been used to assess changes in a neurotransmitter -glutamate (Glu)- in both brain and spinal cord and has shown potential in a number of diseases including Alzheimer’s-like dementia, Parkinsonism and Huntington’s Disease and Motor neuron diseases.

Sodium ions play a central role in membrane transport and cell homeostasis. Increased sodium concentration has been observed in brain tumors as well as neurodegenerative diseases including Alzheimer’s disease, multiple sclerosis and Huntington’s disease. While 23Na MRI of the human brain was first performed over 20 years ago, the low concentration of 23Na compared to 1H and rapid T2 decay resulted in low signal to noise (SNR) and long acquisition times, limiting its diagnostic feasibility. Recent advances in MR technology including the move to higher field strengths (e.g. 9.4T) and ultra-short echo-time sequences, has rekindled interest in 23Na imaging. 

The glymphatic pathway has been proposed as a key contributor to the clearance of fluid and metabolic waste products, such as amyloid beta and tau, from the brain. Recently, dynamic contrast-enhanced MRI has been used to visualize the glymphatic system and monitor CSF-interstitial fluid exchange in normal and Type 2 diabetes mellitus rats, with the latter showing impaired clearance of interstitial fluid. It has also been proposed that glymphatic function may be compromised in motor neuron disease (MND) patients. 

In vitro fertilisation (IVF) is a process of fertilisation where an egg is combined with sperm outside the female body, in vitro ("in glass"). The process involves monitoring and stimulating a person's ovulatory process, removing an ovum or ova (egg or eggs) from their ovaries and letting sperm fertilise them in a culture medium in a laboratory. After the fertilised egg undergoes embryo culture for 2–6 days, it is implanted in the same or another person's uterus, with the intention of establishing a successful pregnancy [1].

Background
Since the first report of successful clinical pregnancy with the use of IVF, several patients
characteristics such as female/male age [2], type of infertility, body mass index (BMI) [3] as well as embryology characteristics like embryo quality are known to play a critical role in clinical outcomes and patient management in IVF [4]. For instance, embryo quality grading systems have been developed based on embryo morphology showing significantly improved clinical outcomes [5]. However, their accuracy remains limited and does not consider individual clinical characteristics of each couple, with no improvements in methodologies applied to date. Indeed, in women younger than 35 years old, 60% of fresh embryos transferred do not result in clinical pregnancy. This failure even rises with increased female age to 75% for women 41–42 years old, and to 85-90% for those who are 43–44 years old [6, 7]. These outcomes are consistent with the fact that normal embryos, as assessed morphologically, fail to implant because of unknown reasons that could be explained due to a lack of integration between clinical characteristics and clinical embryology features; on the other hand, morphologically abnormal embryos could still
result in healthy live births [7]. For these reasons, developing reliable methodologies that integrate clinical characteristics as well as clinical embryology features to assist the clinical decisions of clinicians as well as scientists through Artificial Intelligence (AI) could revolutionise the management of patients undergoing assisted reproductive technologies (ART).

Clinical Decision Support Systems (CDSS) are becoming an increasingly common tool for helping clinicians make better care decisions for their patients in many areas of medicine. At present, significant development has seen value in radiology and pathology disciplines, with some groups developing CDSS for physician problems as well as AI technology becoming more advanced. Indeed, it is anticipated that future medical practitioners will be working alongside these systems in every specialty [8]. However, in the field of IVF that future is under development and heavily reliant only on digital information from images and time-lapse videos of embryos [9] without integrating clinical information that identifies real treatment planning and prognosis that can potentially improve the chances of achieving pregnancy.

This project aims to use the latest developments in AI, machine learning and deep learning in the field of IVF by testing and implementing a more achievable technology that is cost and time-efficient. The proposed system will calculate the probability of pregnancy for each embryo after analysing the image of the embryo whilst considering confounding clinical factors.

Blockchain technology and its popular cryptocurrencies such as bitcoin and Ethereum have most revolutionary technological advances in recent history, capable of transforming businesses, government, and social interactions. However, there is a darker side to this technology which is the immense energy consumption and potential climate impact of the blockchain and cryptocurrencies. According to Digiconomist's Bitcoin Energy Consumption Index, Bitcoin has such an impact on the environment (118.9TWh/year) – considering the huge amount of energy used – that it compares to the power consumption of countries like the Netherlands (117.1 TWh/year) or Pakistan (125.9TWh/year). There is an urgent need to address this danger to ensure the long-term sustainability of IoT.

Cloud Data centres are designed to support the business requirements of cloud clients. However, due to the complexities of data centre infrastructure and their software systems, cloud service providers often do not have access to quality data regarding their IT equipment. This hinders their ability to better optimise the quality of their services and system performance. A clear message from across the industry is that better data allows for better decision making and resource management. A digital twin of a data centre is a virtual or digitised replica of a data centre that would allow real-time analysis of monitoring data of the data centre to avoid problems before they occur, prevent downtime, develop new opportunities such as energy-saving and even plan for the future by using simulations. With the explosion of Internet-of-Things (IoT) sensors such as power, humidity and temperature and the advancement of software and monitoring systems in cloud data centres, building data centres digital twins that would allow for better resource management and enhancement of data centre efficiency is now possible.

Pathfinding is fundamental operation in video game AI: virtual characters need to move from location A to location B in order to explore their environment, gather resources or otherwise coordinate themselves in the course of play. Though simple in principle such problems are surprisingly challenging for game developers: paths should be short and appear realistic but they must be computed very quickly, usually with limited CPU resources and using only small amounts of memory.

As a pregnancy approaches term (the point at which the foetus is considered fully developed), decisions are made about the timing of birth and the way babies are born. These decisions are incredibly challenging for clinicians and pregnant women. Digital health records, advances in big data, machine learning and artificial intelligence methodologies, and novel data visualisation capabilities have opened up opportunities for a dynamic, ‘Learning Health System’ – where data can be harnessed to inform real-time and personalised decision-making. Existing linked administrative databases already capture Australian women and children's observed birth events and their actual health and well-being outcomes. The latest machine learning and artificial intelligence advancements can mine these datasets to create prediction models that can forecast the likely outcomes of current practices.

Traumatic brain injury (TBI) is an injury to the brain caused by an external force from incidents such as motor vehicle crashes, falls, assault or sports collisions. Almost seventy million individuals globally are estimated to suffer from TBI per annum [1], deeming it a major public health concern which is estimated to cost the global economy approximately $US400 billion annually [2]. Early identification of severe TBI with proper assessment and treatment lowers the risk of secondary injury and subsequent long-term disability and subsequent costs. Missed diagnoses can lead to severe complications, consequences and increased cost.

Deep learning approaches on CT have been gaining popularity. Pretrained convolutional neural networks have been recently used to detect COVID-19 [3] and hybrid methods are evolving that combine template matching, artificial neural networks and active contours for segmentation of significant anatomical landmarks and estimation of haematoma volume on brain CT scans [5]. Improving deep learning algorithms to accurately identify and classify head CT scan abnormalities have been identified as requiring urgent attention [4], opening up the possibility to translate into automating the triage process.

Pose Tracking is the task of estimating multi-person human poses in videos and assigning unique instance IDs for each keypoint across frames. Accurate estimation of human keypoint-trajectories is useful for human action recognition, human interaction understanding, motion capture and animation.

Deep learning has achieved ground-breaking performance in many vision tasks in the recent years. The objective of this project is to apply the state-of-the-art deep learning based image classification/detection networks such as ResNet or Faster RCNN for classifying CT or X-Ray images.

This is a "research project" best for students who are independent and willing to take up challenges with high expectation in the grade when fulfilled the somewhat challenging requirements. Under-performing is likely to fail to obtain the passing requirements. It is also a good practice for students who wish to pursue further study at a postgraduate/PhD level.

Neuroscience is becoming an exciting and multidisciplinary field, with a combination of biology, psychology, engineering, and large-data processing. This project is suitable for those who are motivated to apply data-processing skills to biological questions. Our research projects investigate how neural circuits in the mouse brain work during a behavioural task; we visualise neural activity in vivo using advance fluorescent microscopy (two-photon imaging), while filming the behaviour of mice. Animal movements need to be classified accurately and efficiently from the video, using deep learning, and to be linked to neural activity in vivo.

For more information, contact the primary supervisor Dr. Tatsuo Sato <tatsuo.sato@monash.edu>

The last two decades have witnessed a sharp rise in the amount of data available to business, government and science. Data visualisations play a crucial role in exploring and understanding this data. They provide an initial grasp of the data and allow the assessment of findings of data analytics techniques. This reliance on visualisations creates a severe accessibility issue
for blind people (by whom we mean people who cannot use graphics even when magnified).

Guidelines for accessible information provision recommend the use of raised line drawings, called tactile graphics, to show spatial data, such as maps or charts. However, tactile drawings must be explored sequentially using touch, making it difficult to quickly obtain an overview of the graphic. To overcome this, guidelines recommend that tactile graphics come with a textual (braille) description. However, such descriptions are fixed and do not allow the blind reader to ask questions about the graphic or the underlying data.

Despite an increase in the usage of AI models in various domains, the reasoning behind the decisions of complex models may remain unclear to the end-user. Understanding why a model entails specific conclusions is crucial in many domains. A natural example of this need for explainability can be drawn from the use of a medical diagnostic system, where it combines patient history, symptoms and test results in a sophisticated way, estimate the probability that a patient has cancer, and give probabilistic prognoses for different treatment options. In order to accept the result, the patient, the doctor, and even the model developers want to understand (and should be told) why the model supports these conclusions. Unfortunately, decades of bitter experience show that the resulting models and reasoning are often too complex for even domain experts, let alone ordinary users, to understand unaided.

Bayesian Networks (BNs) have become popular as a probabilistic AI tool for working with complex problems, especially in the field of medical and law. A Bayesian network consists of a graph and probability tables, together representing a joint probability distribution. From a Bayesian network, any probability of interest can be computed. The graphical structure contains information about (in)dependencies between variables, which makes a Bayesian network particularly suitable for modelling complex relations between variables.

Expected outcomes: The student will learn inference and representation learning methods for network data. The knowledge can be easily used to analyse other networks, including but not limited to social networks, citation networks, and communication networks. A research publication in a refereed AI conference or journal is expected. A student taking this project should ideally have at least a reasonable background mathematical knowledge, including differential calculus (e.g., partial derivatives) and matrix determinants. The student should also know how to program with either Matlab, Java, or Python. Ideally, the student understands the process of data analysis, which includes data pre-processing, algorithms selection, and evaluation.

 Automation has affected employment at least as far back as Gutenberg, the introduction of the printing press and the effect on scribes and others. Such changes have occurred in the centuries since. In more recent times, we see electronic intelligence showing increasingly rapid advances, with examples including (e.g.) easily accessible, free, rapid and often somewhat reliable language translation. More recent advances include the increasing emergence of driverless cars. This is an area in which rapid changes continue to occur, most recently as the world both deals with and looks to emerge from COVID-19 coronavirus and we seek a sustainable path forward.

  We collect a variety of human opinions to help model and predict the various changes in various parts of the workforce. This will assist in planning, especially as the world looks to re-emerge after - and during - COVID-19.  We follow through on earlier work by Frey and Osborne (2013, 2017) and others in endeavouring to model how automation will affect the future of work.  We extend this now to endeavour to determine with the new issues of how COVID-19 coronavirus will affect the future of work and how societies, economies and workforces will respond to issues of climate change and the likely regular occurrence of bushfires such as those of the Australian 2019-2020 summer.

  We will survey a variety of people from various backgrounds in order to find their views on matters such as these - which jobs will continue as they are, be partly automated and be totally automated. We consider approximately 683 individual occupations, covering a majority of the U.S. workforce. We then consider various attributes of these occupations, as given by the Occupational Information Network (O*NET) data-base. Using a subset of these occupations, we survey a group of experts to predict (probabilistically) whether these occupations will be automated, augmented or unaffected by emerging technologies. Using this data, a classification algorithm is then trained to predict the probability that an occupation in the data-set is automated, augmented or simply unchanged at some in the future. Using the O*NET attributes and the opinions of various surveyed humans, we then endeavour to model and predict which jobs and occupations will be partly augmented or totally automated, and also - for those occupations that are predicted to be augmented or totally automated - which technologies will affect them.

It is our anticipation that the work will commence with, in parallel, the survey for collecting the data and a comparison of machine learning methods on artificial pseudo-randomly generated data. The world will see all too many changes during and in the aftermath of COVID-19. We hope this project - with its multidisciplinary team - to be one of the early projects anticipating where job markets might head.

Theory and applications in data analytics of time series became popular in the past few years due to the availability of data in various sources. This project aims to investigate and generalise Hybrid and Neural Network methods in time series to develop forecast algorithms. The methodology will be developed as a theoretical construct together with wide variety of applications.

    DNA or RNA motif discovery is a popular biological method to identify over-represented DNA or RNA sequences in next generation sequencing experiments. These motifs represent the binding site of transcription factors or RNA-binding proteins. DNA or RNA binding sites are often variable. However, all motif discovery tools report redundant motifs that poorly represent the biological variability of the same motif, hence renders the identification of the binding protein difficult. Here we propose for the first time to apply the Bayesian information-theoretic Minimum Message Length (MML) principle to optimise the clustering of over-represented DNA or RNA motifs in order to predict binding sites that are biologically relevant.

    This will be used primarily for regenerative medicine.

    Minimum Message Length (MML) (Wallace and Boulton, 1968; Wallace and Dowe, 1999a; Wallace, 2005) is a Bayesian information-theoretic principle in machine learning, statistics and data science.  MML can be thought of in different ways - it is like Ockham's razor, seeking a simple theory that fits the data well. It can also be thought of as file compression - where data has structure, it is more likely to compress, and the greater the structure the more it should compress.

The relationship (in principle) between MML and Solomonoff-Kolmogorov (Wallace and Dowe, 1999a) means that MML can, given sufficient data and sufficient search time, infer arbitrarily closely to any model underlying data.

Develop NLP tools to track politicians’ campaign promises on traditional and social media: With applications to Australian, Indian and/or US politics.

This project is an opportunity for an excellent student to develop and demonstrate expertise in NLP. The project will develop tools that can track the extent to which and ways in which politicians’ campaign promises are disseminated through traditional and social media during election campaigns. For selected Australian, Indian and/or US election campaigns, we will identify the campaign promises of each of the main parties and candidates from their election manifestos, platforms and official websites. The NLP tools to be developed and applied in this project will then track these campaign promises in traditional and social media. The analysis of traditional media will focus on digitally archived national and regional newspapers, while the analysis of social media will focus on Twitter data. The NLP tools to be deployed in this project will identify the frequency and contexts in which different campaign promises receive attention. These tools will also quantify the ways in which campaign promises are featured, for instance whether the promises are referred to positively or negatively by candidates and citizens.

This project is of significant academic and practical relevance, and we expect it to lead to one or more high-profile international publications on which the student has the opportunity to become a co-author. In terms of NLP research, the project will advance supervised text classification methods, including sentiment analysis, word embeddings and neural networks.

The project is also positioned at the cutting edge of comparative Political Science. It is widely acknowledged that for democracies to function effectively, political candidates must offer voters meaningful choices during election campaigns. This involves parties making campaign promises on policy issues, which are communicated to voters through traditional and social media. This project therefore examines processes that lie are at the heart of democratic practice. The elections that may be the focus of the applications include the 2019 Australian national election, the 2019 Indian general election and the 2016 and/or 2020 US Presidential elections.

Required skills

A foundation in NLP, supervised text classification methods, including sentiment analysis, word embeddings and neural networks. An affinity with the politics of India and/or the US.

The supervisory team

The student will be supervised by Yuan-Fang Li in the Faculty of IT.
Prof. Robert Thomson in the School of Social Sciences will provide supervision on the Political Science aspects of the project.

International collaborations

The student will become part of an interdisciplinary and international network of prominent researchers in this field. The project is part of the University of California San Diego-Monash joint project "AI for Stronger Democracy and Policy Performance", which was initiated by Mark Andrejevic (Monash School of Communications), Wray Buntine (Monash Faculty of IT), Seth Hill (UCSD, Political Science), Ndapa Nagashole (UCSD, Computer Science), Christina Schneider (UCSD, Political Science), and Robert Thomson (Monash, Social Sciences).

The project is also linked to an international network of Social Scientists and Data Scientists led by Profs. Elin Naurin (University of Gothenburg) and Robert Thomson (Monash), funded by the Bank of Sweden. This network is developing both qualitative and quantitative automated methods for analysing what political parties promise to voters during election campaigns:
https://www.rj.se/en/anslag/2019/mixed-methods-for-analyzing-political-parties-promises-to-voters-during-election-campaigns/

In addition, the project is linked to an ongoing established Political Science project on campaign promises: the Comparative Party Pledges Project, which is co-led by Professors Naurin (Gothenburg), Royed (Alabama) and Thomson (Monash). A summary of some of the work of this project can be found here:
https://ajps.org/2017/06/07/the-fulfillment-of-parties-election-pledges-a-comparative-study-on-the-impact-of-power-sharing/

Issues and solutions exist on different aspects of the management of real-time data, such as persistence, visualisation, and online processing. This project is a research project to identify the significant issues of real-time data management in structural health monitoring (SHM), particularly for bridges, and implement an integrated software solution for enterprise usage. This project involves time series database design, visualisation and online processing of time series, and service-oriented and web-based software development.

Specific scope of works and deliverable are:

1.       A survey of issues on managing the real-time monitoring data of bridges,

2.       An integrated solution of real-time management, resolving the identified issues,

3.       A software system as a reference implementation of the proposing real-time data management solution,

4.       The project deliverable involves software engineering design, coding and testing.

Please note that this is a minor thesis project for units FIT5126, FIT5127 and FIT5128 on Clayton campus. Besides the project deliverables, the student also needs to achieve the assessment standard for these units. For the details, please check https://www.monash.edu/it/current-students/enrolment/honours-and-minor-thesis.

This project focuses on the locomotion pattern of freely moving animals. The model organism we used is C. elegans, a transparent nematode about 1 mm, which displays a sinusoidal movement on the plates. The locomotor performance of adult worms can be decomposed into four fundamental patterns called eigenworms using principal component analysis, and the locomotion of freely moving worms exploring the environment is represented as a linear combination of these four patterns (Ref: https://doi.org/10.1371/journal.pcbi.1000028 and https://doi.org/10.1073/pnas.1211447110  ). Using a commercial system, Wormlab, we successfully tracked the locomotion in different developmental stages in male and female animals after hatch. Further quantitative characterization on different locomotion parameter, including locomotion speed, reversal rate and bending angle, will allow us to precisely evaluate the animal's locomotion and the effect of distinctive environmental cues in animals locomotion. 

For more information, contact the primary supervisor Dr. Jie Liu <Jie.Liu1@monash.edu>

A major challenge in cancer therapeutics is to kill tumour cells without harming normal cells in the body. Traditional chemotherapy tries to do this by killing cells that are fast dividing, a characteristic hallmark of cancer cells, however as many other cells in the body are also fast dividing – such as those in the hair and the gut – chemotherapy typically results in undesirable side effects. Newer targeted therapies are designed to specifically target cancer cells, by exploiting the genetic changes that distinguish tumour cells from normal cells. One emerging and exciting concept for the development of targeted therapies is known as ‘synthetic lethality’ – whereby the function of gene X only becomes essential if gene Y is mutated. In this case, inhibiting gene X would only kill cancer cells (having mutated Y) without affecting normal cells. Research in Dr Lan Nguyen’s lab have developed preliminary bioinformatics approaches to identify these synthetic lethal X/Y pairs. This project will build on these work to identify potent synthetic lethal gene pairs for breast cancer (and others), based on which new effective targeted therapies could be developed.

Candidate students will work with real-life high-throughput molecular and biomedical data available in the Nguyen lab under supervision of experienced researchers. They will utilise key data science techniques including data processing, integration, analysis and visualisation.

Note on project - Preferably 2 semesters, but is currently designed for 1 semester.

For more information, contact the primary supervisor Dr. Lan Nguyen <lan.k.nguyen@monash.edu>

Despite enormous progress in research, cancer remains a devastating disease worldwide. Since generally not all patients will respond to a specific therapy, a great challenge in cancer treatment is the ability to predict which patients would benefit (or not) to a therapy of choice. This helps improve treatment efficacy and minimise unnecessary sufferings by non-responders. There is thus a pressing need to identify robust biomarkers (i.e. genes/proteins) that can accurately predict the right patients for the right drugs. With the increasing availability of molecular and drug-response data, machine learning approaches provide a powerful tool for this task. This project will utilise key data science techniques including data processing, integration, analysis and visualisation; and then use these data to develop useful machine learning models to identify optimal biomarkers for different cancer types. This work will build upon a Support Vector Machine-based pipeline in the Nguyen lab.

Students will have a unique opportunity to work with real-life biomedical data and contribute towards solving a critical challenge in cancer research in Dr Lan Nguyen’s lab. They will be able to develop their skills in a highly interdisciplinary research group with expertise in both computational & biomedical fields. There are also potential opportunities to continue to PhD studies.

Note on project - Preferably 2 semesters, but the project is currently designed for 1 semester.

For more information, contact the primary supervisor Dr Lan Nguyen <lan.k.nguyen@monash.edu>

Antimicrobial resistance (AMR) continues to evolve as a major threat to human health and new strategies are required for the treatment of AMR infections. Bacteriophages (phages) that kill bacterial pathogens are being identified for use in phage therapies, with the intention to apply these bactericidal viruses directly into the infection sites in bespoke phage cocktails. Using such a biological agent for infection control requires deep understanding of the phage. Thus, and despite the great unsampled phage diversity for this purpose, an critical issue hampering the roll out of phage therapy is the poor-quality annotation of many of the phage genomes.
To this end, our lab has developed machine learning based toolkits to accurately predict different types of phage proteins, including anti-CRISPR proteins using the PaCRISPR web server (https://pacrispr.erc.monash.edu/, published at Nucleic Acids Research) and phage structural proteins using our STEP3 predictor (https://step3.erc.monash.edu/, under revision at mSystems). We also developed a comprehensive platform AcrHub (https://pacrispr.erc.monash.edu/AcrHub/, published at Nucleic Acids Research) to provide all-in-one service for comprehensively categorizing, predicting and visualizing the anti-CRISPR proteins of phages.
This project has three aims to further our current work: 1) expanding STEP3 to predict the structural proteins into subtypes, such as those in the Capsid, Neck and Tail parts of phage particles; 2) expanding PaCRISPR to divide the non-structural proteins into different subtypes, such as anti-CRISPRs and various types of phage enzymes; and 3) building a platform to make annotations for phage genomes through combining work from 1) and 2).

This project can be taken as a single or double semester research. 

• Single semester (Aim 1)
• Double semester 2 (Aims 1, 2 and 3) 

For more information, contact the primary supervisor Prof. Trevor Lithgow <trevor.lithgow@monash.edu>

Multidrug resistance (MDR) poses critical challenges to global health. In 2017 the World Health Organization identified Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacteriaceae as the top-priority pathogens that urgently require development of novel therapeutic options. Recently, bacteriophage therapy has attracted extensive attention owing to its potential of being used as novel antimicrobials to combat MDR pathogens.

Bacteriophages are the most abundant and diverse organisms in the biosphere. To date, there are approximately 10,000 complete phage genomes in the GenBank database. Despite their relatively small genomes, bacteriophages show impressive genomic diversity and complex evolutionary relationships that do not follow traditional hierarchical phylogeny. The limited understanding of phage genomes and evolutionary relationship has significantly hindered the development and optimisation of effective bacteriophage therapy against ‘superbugs’.

This project will conduct comparative genomic and evolutionary analyses of bacteriophages infecting top-priority Gram-negative pathogens. The available bacteriophage genomes from public databases will be combined with our own sequenced genomes for comprehensive analysis. Evolutionary analysis for core genes will be conducted and compared with the phylogeny of host bacteria to investigate the bacteriophage-bacteria interactions in the context of coevolution.

For more information, contact the primary supervisor Prof Jian Li <jian.li@monash.edu>

Neural module networks (NMNs) [1] support explainable question answering over text [2] by parsing a natural-language question into a program. Such a program consists of a number of differentiable neural modules that can be executed on text in a soft way, operating over attention scores. As a result, NMNs learn to jointly program and execute these programs in an end-to-end way.

Designed mainly for compositional (multi-hop) reasoning, NMNs provide limited support for numerical reasoning. It only supports number comparison, finding min/max in a paragraph and date differences.

The performance of deep neural models rely on large amounts of labeled data, however, most data remain unlabeled in the real world scenario. While annotating data is expensive and time consuming, active learning seeks to choose the most appropriate and worthwhile data for human annotation. It is noticed that humans give labels to some specific data with some labeling reasons or rationales,  which are often existing in the data.  The goal of this research is to develop effective deep active learning techniques with rationales.

Mission-critical systems have to comply to various formal standards – e.g. DO-178C and ISO26262 - about their operation, usually heavily relying on formal specification languages such as TLA+. This presents many challenges to developers in terms of how to write, read and communicate the target system’s formal specifications. In most cases, having the right formal methods experts to write specifications does not solve the problem as the wider development team needs to be able to deeply understand the formal specifications. This project will explore visualisation techniques to translate systems’ specifications developed in TLA+ into user-friendly diagrams/visualisations that are easy to understand.
 

User reviews on app distribution platforms such as Google Play store and Apple App store are a valuable source of information, ideas, and requests from users. They reflect the needs and challenges users encounter including bugs, feature requests, and design. Recent research has shown that reviews can also serve as a proxy for understanding the values of the users and how users perceive that their values have been violated by the mobile app/mobile app developers. However, there are limited studies that show whether mobile app updates fix violations of the user's values and to what extent. This study aims to systematically analyse the evolution of user reviews with respect to human values while taking into account a set of mobile app updates.

This project will develop new technologies for supervised machine learning from time series building upon our world-leading and award winning research in the area. See my time series research for details of the research program on which this research will build.

The world is dynamic, in constant flux. However, machine learning typically learns static models from historical data. As the world changes, these models decline in performance, sometimes catastrophically so. This PhD will develop technologies for addressing this serious problem, building upon our groundbreaking research into the problem.

Bacteria can live in almost all possible environments on earth. In general, they contribute to the stability and health of ecosystems and are very beneficial. However, some bacteria when in contact with humans can cause diseases. Despite the efforts to control them using antimicrobial agents, some of these bacteria have developed resistance and impose a threat to public health. The ability to resist antimicrobial agents lies on the genetic content of these bacteria, in their genes. Though this is a negative outcome for humans, antimicrobial resistance (AMR) genes give bacteria higher chances of survival and are likely also present in environmental bacteria. However, little is known of the origin of these genes and their prevalence in other ecosystems (e.g. Antarctic, not dominated by humans).

This project aims to evaluate the genetic content of bacteria of different environmental samples and assess the prevalence and origin of AMR genes. Methods from metagenomics, comparative genomics and phylogenetics will be used for this project.

For more information, contact the primary supervisor A/Prof Chris Greening <chris.greening@monash.edu>

Proteomics data generated by cutting-edge mass spectrometers play a crucial part in early disease diagnosis, prognosis and drug development in the biomedical sector. It can be used to understand the expression, structure, function, interactions and modifications of virtually any protein in any cell, tissue or organ. Moreover, proteomics can be used in conjunction with other “omics” technologies such as genomics, transcriptomics or metabolomics to further unravel the complexity of signalling pathways and other subcellular systems.

Although the majority of proteomics data is visualised and reported for convenience of the end user on the protein level, the mass spectrometer is actually measuring and quantifying protein fragments (peptides), which are then “converted” computationally into protein-centric data. This conversion from peptide to protein level is quite error-prone, and it is better – and often even required - to present proteomics data on the peptide level and not on the protein level. However, appropriate visualisation strategies and comparative analyses are not well established yet.

This project will focus putting computational algorithms and statistical strategies in place to enable a peptide-centric analysis when interrogating and visualising proteomics data .

The student will be required to use programming languages such as R and/or Python to interrogate, manipulate, analyse and visualize proteomic datasets and to implement these methodologies within the Monash Proteomics and Metabolomics facility. They will get hands-on experience with real-world biomedical data, and learn critical data science skills including programming, data manipulation, statistical analysis and reporting.

For more information, contact the primary supervisor A/Prof Ralf Schittenhelm <ralf.schittenhelm@monash.edu>

Lipids such as cholesterol or triglycerides are involved in a plethora of medical disorders and diseases ranging from cardiovascular diseases (including obesity and artherosclerosis) to neurodegenerative disorders such as Parkinson’s disease. An in-depth analysis of individual lipid classes and species is often indispensable to unravel the mechanisms underlying disease onset and progression.

Lipidomics, which is the large-scale study of cellular lipids in biological systems using mass spectrometry, is the only technique which is able to provide this critical information. However, there is still a lack of accessible tools for the interpretation and downstream analysis of these complex lipidomics datasets.

In this project we are planning to develop a freely available, interactive and easy-to-use web-based prototype system for lipidomics analysis. 

For more information, contact the primary supervisor A/Prof Ralf Schittenhelm <ralf.schittenhelm@monash.edu>

DeepLabCut™ is an efficient method for 3D markerless pose estimation based on transfer learning with deep neural networks that achieves excellent results across a broad collection of behaviours. This project will utilise the DeepLabCut package to analyse the behaviour of rats and mice as they are trained and tested on reward-based learning tasks designed to examine aspects of attention, memory and impulsive behaviour. These tasks are performed in touchscreen chambers that have on one side a monitor to display visual images in different arrangements and on the other side, a sugar pellet magazine for the dispensing of rewards. Importantly, these chambers are attached, via an automated sorting mechanism, to a home cage in which rats and mice live in social groups. They access the touchscreen tests in a self-directed manner, without experimenter intervention, so that their behaviour is as naturalistic as possible for a laboratory testing environment.
Video recordings of test sessions will be examined to determine patterns in behaviour that correlate with learning. These include how the speed of movement between the stimulus on screen and reward collection changes over time, side-preference patterns for turning, and types of movement in anticipation of an expected outcome (side-to-side head movements that scan for potential stimulus presentation location). Videos are recorded during different times of the day and night, therefore, information already collected about when learning is fastest will also inform the assessment of movement and behaviour. 

For more information contact the primary supervisor Dr Claire Foldi <claire.foldi@monash.edu>

Activity and movement are fundamental diagnostic parameters of animal behaviour. However, measuring long-term individual movement within groups was not possible until recently. Our ActivityMonitor provides accurate individual movement data in a fully automated way. This is a unique solution for the 24/7 long-term tracking of individual animals living in groups, which utilises an array of RFID readers positioned under the home cage of rats and mice that are implanted with RFID transponders. The AcitivtyMonitor is connected via an automated sorting mechanism to a touchscreen testing chamber, in which we present images on a screen that are paired with reward-based outomes and result in the delivery of sugar pellets. With touschreen testing we can study various aspects of reward-learning and cognitive behaviour and, combined with RFID technology, we reveal complexity in individual learning styles. It is also the case that learning changes an animal’s behaviour, which, in turn can change the behaviour of the social group. This system therefore has the capacity to inform us about how feedback from reinforcement learning changes social and cognitive behaviour over time.   
By analysing the data output from ActivityMonitor, this project will provide answers to the following questions:
1. How do rats and mice behave in the home-cage, before, during and after cognitive training?
2. How does this behaviour change, depending on the time of day?
3. Can we identify, through RFID movement data, a social heirarchy in group-living rats and mice?

For more information, contact the primary supervisor Dr Claire Foldi <claire.foldi@monash.edu>

Working remotely under the COVID-19 pandemic has given rise to the demand for cloud-based technology, including online file sharing and cloud storage services. However, attackers have recently abused these platforms and propagated the emails that contain a file-sharing link to bypass the email filter. A typical example is that criminals can easily create and share phishing forms through legitimate form builders, e.g., Google Form to trick users into handing over sensitive information such as password or credit card number. This attack is hard to be detected by traditional detection schemes since there is no malicious content delivered in the email, and the links can be created quickly.

The objective of this project is to analyse the emerging phishing attacks and automatically generate security contents for file sharing links, e.g., warnings or indicators to help users understand the attacks and identify phishing emails before getting harmed. A prototype would be developed to be integrated into the major email clients, e.g., Outlook, Gmail.

People are continuously receiving unsolicited emails where phishers impersonate legitimate organisations or trusted sender to harvest victim credentials. The rapid advance of AI boosts recent automatic detection of phishing attempts but also provides hackers with the opportunities to build increasingly sophisticated phishing tactics to bypass the filter. While attackers leverage social engineering to exploit human weakness, human skills can be a powerful component in cyber defence such as cognitive function and professional judgment.

The project aims to develop a system that incorporates human intelligence and AI to combat phishing attacks. The goal is to design a collaborative way that human strengths and AI harness, extend, and complement the capability of each other. This will build a sense of responsibility and trust for users and maximise the cyber resilience against phishing attacks.

Deep learning has achieved ground-breaking performance in many 2D vision tasks in the recent years. With more and more 3D data available such as those captured by Lidar, the next research trend is doing advanced perception on 3D data. The objective of this project is to study the state-of-the-art object detection techniques for 3D point clouds such as PointNet and PointVoxel.

This is a "research project" best for students who are independent and willing to take up challenges with high expectation in the grade when fulfilled the somewhat challenging requirements. Under-performing is likely to fail to obtain the passing requirements. It is also a good practice for students who wish to pursue further study at a postgraduate/PhD level.

Modern machine learning is increasingly applied to create amazing new technologies and user experiences, many of which involve training machines to learn responsibly from sensitive data, such as personal photos or email. Ideally, the parameters of trained machine-learning models should encode general patterns rather than facts about specific training examples.

It is quite challenging to access to videos for people who are blind or have low vision (BLV), particularly creating audio descriptions that describe the scenes without interfering the dialogues in a video. There is also the challenge of providing additional information using multi-modal feedback, that is using non-speech audio and haptics.

In this project, you will work on analysing videos using deep learning frameworks and extract information to generate a multi-modal interaction that includes audio descriptions, non-speech audio, and haptics. The input videos will include online lectures, movie clips, and media artworks and the output feedback will be audio descriptions that describes the background and foreground objects in the frames as well as the mood of people in these frames.

Haptic ring is a wearable device that is used by people who are blind or have low vision. It provides electro-vibration feedback on different locations of users' fingers. Its primary use is to extend the user interaction with touch screens in which haptic feedback is restricted due to battery consumptions. 

In this project you will work on improving the haptic ring by investigating design improvements that will allow the device to provide more accurate feedback. We will provide support for the design and programming components of the project.

People who are blind or have low vision (BLV) access documents using screen readers such as JAWS and NVDA. These screen readers emulates a cursor moving around the screen using arrow keys or various shortcut combinations. However, this way of interaction is vey slow and not ideal for getting an overview of a document and navigating to relevant sections.

This project will investigate novel ways of improving document access by extending existing screen reader features and document editing softwares such as Microsoft Word, so that BLV can quickly get an overview of a document as well as navigation cues for important sections.

During the project you will first conduct surveys to understand the current state of access to various documents used in the workplace. Second, you will focus on improving the NVDA (https://github.com/nvaccess/nvda/) screen reader. Third, you will improve the accessibility of Word documents itself using Microsoft VBA (Visual Basic of Application) and/or Microsoft Visual Studio Code (https://github.com/microsoft/vscode).

Since the 1990s, researchers have known that commonly-used public-key cryptosystems (such as RSA and Diffie-Hellman systems) could be potentially broken using efficient algorithms running on a special type of computer based on the principles of quantum mechanics, known as a quantum computer. Due to significant recent advances in quantum computing technology, this threat may become a practical reality in the coming years. To mitigate against this threat, new `post-quantum’ (a.k.a. `quantum-resistant’) algorithm standards for public-key cryptography are in development [1], that are believed to be resistant against quantum computing attacks.

With the rise of software systems ranging from personal assistance to the nation's facilities, software defects become more critical concerns as they can cost millions of dollars as well as impact human lives. Yet, at the breakneck pace of rapid software development settings (like CI/CD, Agile, Rapid Releases), Software Quality Assurance (QA) practices (e.g., code review and software testing) nowadays are still time-consuming.

In this project you will work on creating a 3D printed platform used with an iPad for people who are blind or have low vision. The platform will allow people to program in the Scratch visual programming language (https://scratch.mit.edu/) using 3D printed blocks. You will program Arduino boards, print models using 3D printers, and integrate these models with an iPad.

Accessing maps is a very challenging task for people with vision impairment. Particularly, navigating a map using panning and zooming and finding information on the screen. 

In this project, you will work on understanding the important sections of a map, and finding the best ways to convey these information to people who are blind or have low vision. To do this, you will use Eye Tracking devices to track sighted people and determine the focused areas on a map. Then, you will use this knowledge to generate accessible maps for blind people.

People who are blind need to touch surfaces and materials to get information. These surfaces can be a Braille paper that has Braille text, a swell paper that has embossed shapes, and a button that is used to turn on and off a device like a TV or to open a train carriage door. Due to Covid-19, hygienic practices will be more and more important, and it will be very restrictive for people who are blind to touch surfaces in public places. Mid-air haptics interfaces can potentially be used in public places or in community areas to provide information and to operate devices without touching a surface.

In this study, we want to investigate the use of mid-air haptics to provide visual information. Mid-air haptics interfaces use an array of tiny speakers to generate ultrasonic waves up in the air. These waves can be programmed in a way that it can create acoustic pressure areas that can be sensed as an air flow on the palm or finger tips.

For Australians with impaired vision, accessible books are a lifeline to education and vital everyday information, and also to the independence and personal autonomy that sighted people take for granted. Yet much literature remains in an inaccessible format. 

One part of the problem is the conversion of textual and graphical information in existing books. This requires algorithms to access the raw content and convert them into an accessible format. However, accessing copyrighted material poses challenges for both the publishers and the consumers. On one hand publishers need to deal with generating alternative content and the costs associated with it. On the other hand, the consumers - in this case people who are blind or have low vision (BLV) - need to navigate the complexities of linking the publishers with accessible content transcribers to get their preferred accessible format such as Braille, or text and audio descriptions.

In this project you will be investigating the best ways of accessing the raw data from book publishers to content transcribers and then to BLV. You will prepare surveys and conduct interviews with publishers, transcribers and BLV to understand the practical challenges that arise in this space. You will then investigate how the results of these interviews and surveys can be integrated with existing frameworks such as the Australian Copyright Amendment (Disability Access and Other Measures) Bill 2017 and the (international) Marrakesh Treaty. The goal is to work out a practical process that also aligns with relevant legislative imperatives. 

Today's society and its organisations are becoming ever-increasingly dependent upon Information and Communication Technology (ICT) systems mostly based in cloud data centres. These cloud data centres, serving as infrastructure for hosting ICT services, are consuming a large amount of electricity leading to (a) high operational costs and (b) high carbon footprint on the environment. In response to these concerns, renewable energy systems are shown to be extremely useful both in reducing dependence on finite fossil fuels and decreasing environmental impacts. However, powering data centres entirely with renewable energy sources such as solar or wind is challenging as they are non-dispatchable and not always available due to their fluctuating nature. Recently, container solutions such as Docker and container orchestration platforms such as Kubernetes, Docker Swarm, or Apache Mesos are gaining increasing use in cloud production environments. Containers provide a lightweight and flexible deployment environment with performance isolation and fine-grained resource sharing to run applications in clouds. This project intends to develop scheduling and auto-scaling algorithms for container orchestration within clouds based on the availability of renewable energy.

BARD: Bayesian Argumentation via Delphi [1] is a software system designed to help groups of intelligence analysts make better decisions. The software was funded by IARPA as part of the larger Crowdsourcing Evidence, Argumentation, Thinking and Evaluation (CREATE) program. The tool, developed at Monash University, uses causal Bayesian networks as underlying structured representations for argument analysis. It uses automated Delphi methods to help groups of analysts develop, improve and present their analyses.

Point of sales (POS) is the data that is recorded at the retailer level when consumers purchase the products.  POS data is becoming increasingly popular for companies to predict their sales. In a supply chain, the POS data are often used by retailers to predict their sales, however, manufacturers and suppliers have not benefited enough from POS data. Retailers place their orders to suppliers as they predict their sales.

The goal is to predict the supplier's demand. We can use either the retailer's orders or POS data to forecast suppliers' demand. The idea is to use Granger's causality​​ to forecast ​one time series​ (suppliers demand)​ from ​another​ time series​ (POS data)​. This is an empirical study and we will be using real sales time series​​ of food company.

Demand forecasting is the basis for a lot of managerial decisions in companies. During the last four decades, researchers and practitioners have developed numerous quantitative and qualitative demand forecasting models including statistical, machine learning, judgmental, and simulation methods. Several endogenous and exogenous variables can influence the dynamics of demand, and hence a single statistical model that only consists of historical sales data is often insufficient to produce accurate forecasts. In practice, the forecasts generated by baseline statistical models are often judgmentally adjusted by forecasters to incorporate factors and information that are not incorporated in the baseline models.

 Using expert judgment in complement to statistically analyzing large amounts of data has been shown to be beneficial for improving forecast accuracy. Moreover, evidence suggests that the human input to forecasts can be improved by providing a systematic approach to structure the information utilized when imposing judgment to make adjustments. However, when the number of peripheral contextual information increases, human factors can hinder judgment for reasons such as personal or social biases, heuristics, cognitive limitations, and system neglect.

 The question is more on how to appropriately process judgment and combine it with statistical/Machine learning forecasts to consistently improve the accuracy of forecasts. 

 Bayesian Networks (BN) are great tools to assess vast amounts of information, analyse their interaction, and combine qualitative with quantitative information for decision making. In this project, we aim to build a BN by combining the contextual information with historical quantitative information in a demand forecasting problem.

 The created BN can act like a Forecasting support system  (FSS) that ease the cognitive burden on the human mind and improves the accuracy of final forecasts by combining qualitative information (human judgment) and quantitative information.

Combating cybercrime and maintaining national security is a global challenge. In light of this the Cybersecurity Capacity Maturity Model for Nations (CMM) has been deployed in over 80 countries; the objective of the CMM is to understand and evaluate cybersecurity capacity within these national contexts in order to support the “well-being, human rights and prosperity”. The outcome of each deployment is a comprehensive report.

Using digital technologies (such as 3D printing, soundscapes, beacon navigation), this project will explore the creation accessible cultural experiences for people who are blind or have low vision (BLV). Accessible materials and space design will be developed and evaluated to provide an evidenced-based framework for producing inclusive experiences with arts and culture, such as Art Galleries. Current partners include the Bendigo Art Gallery and Science Gallery Melbourne.

Within the faculty's Centre for Organisational and Community Informatics, the Archives and the Rights of the Child Research Program is investigating ways to re-imagine recordkeeping systems in support of responsive and accountable child-centred and family focused out-of-home care. Progressive child protection practice recognises the need, where possible, to support and strengthen parental engagement in the system in order to ensure the best interests of the child. 'No single strategy is of itself effective in protecting children. However, the most important factor contributing to success is the quality of the relationship between the child's family and the responsible professional' (Dartington, 1995 quoted in Qld Department of Communities, Child Safety and Disability Services 2013). Child protection and court processes generate a mountain of documentation that can be overwhelming and confusing to navigate, hard to manage and keep track of, especially if parents are also dealing with health and behavioural issues. Being on top of the paperwork handed out by workers, providing the documentation the system demands in a timely fashion and ensuring that records are created to document interactions, etc. could be one way in which child protection outcomes could be improved.

The suitability of a search method for solving an optimisation problem instance depends on the structure of the fitness landscape of that instance. A fitness landscape in the context of combinatorial optimisation problems refers to the search space of possible solutions, the fitness function, and the neighbourhood operator.

Networks are a useful way to model complex data and systems, from biology, to engineering, to social, economic and political structures.  Visualising such network-structured data can be a difficult challenge: when connectivity is high diagrammatic representations of the networks become very dense and tangled.  Immersive environments such as virtual and augmented reality offer a way to use the space around analysts to spread out the network to better show the connectivity.  This project will explore such immersive environments and natural interaction techniques to allow analysts to better explore complex network data.