Both humans and some animals learn on multiple time-scales. If you tell a person a number and ask them to recite it back they will be able to do it. But ask a professional how they learned their craft and they will tell you it took years and years. However, most reinforcement learning algorithms only take into account one time scale, meaning one discount factor for future rewards. This project aims to investigate weather multi-timescale nexting, trying to optimist the immediate future on multiple time-scales, can provide a more in-depth approximation of learning in biological systems.
University of Western Australia
Despite spending more hours at the Western Australian Academy of Performing Arts than he should in a given week, playing anything from piano to trombone to saxophone, David still finds time to be a second-year Mathematics, Computer Science and Data science student at the University of Western Australia. He will often be found saying ‘Yes Maths is actually my first major’ despite helping run a Maker space, Programming Society and Astronomy Club. He loves finding the odd links and connections between fields and often finds that applying simple things from one can make a big difference in another.
The only thing he struggles more with than finding that pesky memory leak is finding enough time to do everything and both of which are a constant never-ending battle. He has a keen interest in Machine learning, specifically computer vision and reinforcement learning as in his words “why not get machines to learn the best way to do something then show us afterwards so we don’t have to” and hopes to apply it wherever he can.