Honours and Masters project

We will design a multi-agent web agent system (powered by LLMs) capable of understanding natural language user preferences, decomposing complex queries into sub-tasks, and dynamically interacting with heterogeneous online tools and databases (e.g., real estate listings, school ranking data, maps). The goal is to generate recommendations that meet users’ multi-objective constraints. Inspired by recent agentic approaches, this project will explore how autonomous web agents can coordinate and perform sequential web-based operations to solve real-world decision-making problems.

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.

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.

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.

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.

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.

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.

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.

This project will investigate counter-unmanned aerial system (C-UAS) technologies for the detection and mitigation of malicious drones. With the increasing accessibility of small UAVs, there is a growing need for effective technical solutions to identify and neutralize unauthorized aerial threats. The project will explore a broad range of C-UAS methods, including but not limited to networking-based detection and coordination techniques, machine learning, and both active and passive mitigation approaches.

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.