I successfully defended my thesis on Friday, June 30, 2023! What an epic conclusion to this amazing eight-year journey. Read more...
Last week at NeurIPS 2021, I presented a paper with Neev Parikh, Omer Gottesman, and George Konidaris, on learning suitable abstract representations for reinforcement learning. We describe novel theoretical conditions for a state abstraction to provably preserve the all-important Markov property, and introduce a practical training objective for approximately learning such an abstraction with deep neural networks---all without requiring reward information, pixel reconstruction, or transition modeling. Read more...
This week at IJCAI 2021, Michael Katz, Tim Klinger, George Konidaris, Matt Riemer, Gerry Tesauro, and I are presenting a paper about making black-box planning more efficient through the discovery and use of macro-actions that intentionally modify a small number of state variables. We describe how to find such "focused macros", and show that they have very intuitive explanations and end up being useful for a wide variety of deterministic planning problems. Read more...
My perspective on the core problems artificial intelligence, viewed through the lens of reinforcement learning. This philosophy guides the majority of my research and helps me to decide on which problems are the most important. Read more...
An essay I wrote in the spring of 2018 on the use of AI for playing board games. It describes the history behind the AI systems that eventually defeated human experts at playing chess, checkers, backgammon, and Go. Read more...