Research

Practice-oriented AI is about bridging the gap between between complex problem domains such as those found in science and research, and AI algorithms and techniques that could be used to solve problems in those domains.

My research focuses on addressing adversarial polcies, a threat to the robustnes of Multi-Agent Reinforecement Learning (MARL).

My current research within the CDT

Currently I am looking at different methods for mitigating an adversarial policy once it has been detected. Previous works have used opponent masking to mitigate adversarial policy, however I have intital results showing that opponent masking underpefroms in certain domains. Based on these results I am now investigating ways of improving opponent masking.

Research by others I’m building on

My current research buids off “A Classification Based Approach to Identifying and Mitigating Adversarial Behaviours in Deep Reinforcement Learning Agents”. My main contributions being an exploration of the current mitigation strategy, and alternative mtitgation startegies which imrpove on it in certain domains.