Research projects in Information Technology

Please note that this PhD topic is offered exclusively at our Monash Malaysia campus and is not available at the Clayton campus.

Core PhD Question

Can Transformer models understand the full lifecycle of a vulnerability; from vulnerable code, to patch, to advisory, to regression risk; and determine whether a security fix is complete, safe, and trustworthy?

So we are not planning to do the following:

Medical VLMs - especially if they can be deployed inside "corporate firewalls" or under the direct governance of health services - potentially offer great advantages over general purpose VLMs for a range of reasons.

However there are fundamental questions over their performance and suitability for deployment inside healthcare, and as to whether they can compete given the mega-infrastructure and ability to upgrade that is available to the big players (OpenAI, Anthropic etc). 

The rapid deployment of increasingly capable AI agents has prompted a fundamental reassessment of how safety should be built into AI systems. Bengio and colleagues have argued that purely agentic training objectives are intrinsically risky and have proposed an alternative paradigm: a non-agentic "Scientist AI" that explains the world from observations rather than acting in it, combining a world model that generates explanatory theories with a question-answering inference machine, and operating with explicit notions of uncertainty so as to mitigate overconfident predictions [1].

Several explanations for hallucination exist, but perhaps the main reason is that they are trained to be plausible.  There is no element of truth-seeking in their construction.  The training content can be filtered to support this, but in many areas truth is not established.   Many text sources are opinions, may contain argumentation and indeed subtle or not so subtle propaganda, written in many dfferent styles, and some reflect misinformed views.   Even Wikipedia sources are known to have bias (so called "establishment bias").

In EdgeVLMOpt (EVO): Optimizing Vision-Language Models for Resource-Constrained Edge Devices, we aim to develop efficient and scalable techniques to enable the deployment of advanced vision-language models (VLMs) on edge hardware. While VLMs have demonstrated strong capabilities in multimodal reasoning and understanding, their high computational and memory demands pose significant challenges for real-time, on-device applications.

In EdgeFusionAI (EFAI): Real-Time Multi-Sensor Multi-Modal Intelligence on Edge Devices, we aim to design and develop efficient techniques for fusing heterogeneous sensory data, including vision, LiDAR, radar, and other modalities, to enable robust and real-time decision-making on resource-constrained edge platforms. This project focuses on building intelligent systems capable of integrating diverse data sources while addressing the challenges of limited computation, memory, and energy availability at the edge.

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 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.

This project aims to develop a computer vision system capable of detecting and classifying domestic geographic landmarks in images and video content. By categorizing locations such as “childcare centre”, “school”, “shopping centre”, “hotel”, “bus stop”, “train station”, and so on, the system provides a reliable way to identify key landmarks in urban and suburban environments. Using advanced machine learning techniques, the model processes visual data to recognize distinct architectural features, signage, and contextual cues associated with each landmark type.

This project focuses on developing a gait detection system leveraging computer vision techniques to recognize individuals in security footage, even when their faces and skin are obscured. Criminals often cover their faces to avoid recognition, making traditional facial recognition unreliable. By analyzing key features of an individual’s gait, such as walking style, speed, and direction, this system provides a robust alternative for visual attribution.