Reinforcement learning (RL) agents have achieved remarkable results on a variety of domains. Although outside the academic sphere their use is more limited than you would think. The main problem is that RL agents, particularly those that utilise neural controllers, come with little to no guarantees.
My current research looks at leveraging results from model-based RL, specifically, world models, with the idea of verifying neural controllers and quantifying uncertainty, to better inform decision making.
I decided to apply to the CDT because, firstly, my research interests, concerns and goals align very closely with the ethos of the CDT and my supervisor. Secondly, the idea of cohort based training was very appealing to me, because for my experience it has built a good sense of community, within which friendships and collaboration can flourish.
