About Me

I am a PhD student based in the Centre for Complexity Science at the University of Warwick, supervised by Professor Matthew Turner. My project has been looking at using the principle which we call future state maximisation as an intrinsic motivation for decision making. This is the idea that, taking everything else to be equal, generally it is preferable to make decisions that maximally keep your options open. The rationale for this is that in an uncertain world preparing yourself to deal with the widest range of potential future scenarios represents an attempt to maximise the amoiunt of control you have over your future environment. This is a principle with potentially wide ranging applicability as it can essentially be applied to any decision making agent in any environment, and intuitively should often lead to "intelligent" behaviour such as the development of reusable skills. Our approach is largely inspired by the empowerment and causal entropic forces frameworks, and our main application has been to a model of collective motion. Here a group of agents move so as to maximise the amount of control they have over the potential future visual states they can access. Simply by each agent following this simple rule a swarm spontaneously emerges which is highly aligned and cohesive with rich dynamics, as well as reproducing a number of features observed in real flocks of starlings. See research for more information.

More generally my research interests are in nonequilibrium statistical mechanics and in recent years I have become very interested in machine learning too, particularly reinforcement learning. In my spare time I have been working on a project where I have been using self-play deep reinforcement learning to train an AI to play a game called "Big 2". This is a four player game of imperfect information with a relatively complex action space, and so represents an interesting challenge. However, the AI is now at the stage where it comfortably beats amateur human players, and has yet to be tested against higher level players. Details of the project can be found on its GitHub page, and I also recently added a paper on the arXiv describing what I did. You can even play against the AI here, although it may take a while to load as I am using a free Heroku node to host it.

Academic Background

Conferences/Summer Schools

You may have seen me at the following: