I am particularly interested in studying four major challenges faced by reinforcement learning agents:
Among other things, I am currently interested in applying recurrent neural networks to tackle these challenges.
I believe that computer games are the ideal testbed for reinforcement learning, since they typically contain well-defined challenging tasks. Success in games should be followed by success in restricted real world environments, as long as state representations encode past events based solely on raw observations, reward signals, and no ad-hoc knowledge. Success in real world environments, however, would still require a wide variety of skills, such as writing an academic webpage.
My previous work has been mostly on the intersection between image analysis, machine learning, and information visualization.
If you want to talk about my academic interests, send an e-mail to pr at this domain.