You can also find my articles on my Google Scholar profile.
We show that with a small amount of labeled data we can train an inverse dynamics model accurate enough to label a huge unlabeled source of online data – here, online videos of people playing Minecraft – from which we can then train a general behavioral prior.
ICML 2021, 2021
An ambitious goal for artificial intelligence is to create agents that behave ethically. Unfortunately, there is widespread disagreement about which ethical theory an agent should follow. This paper translates philosophical work on moral uncertainty into an RL algorithm and investigates how moral uncertainty can avoid extreme behavior compared to single theories.
An important challenge in reinforcement learning is training agents that can solve a wide variety of tasks. If tasks depend on each other (e.g. needing to learn to walk before learning to run), curriculum learning can speed up learning by focusing on the next best task to learn. We explore curriculum learning in a complex, visual domain with many hard exploration challenges: Minecraft.
The definite version of the Go-Explore algorithm. On top of the results from the original pre-print, it introduces a dynamic representation that supports all Atari games, a variant in which the exploration phase can be performed in stochastic environments, and demonstrates Go-Explore working in a robotics environment.
This work presents an exploration and imitation-learning-based agent capable of state-of-the-art performance in playing text-based computer games using the Go-Explore exploration algorithm.
In this paper, we introduce a novel form of value function, Q(s, s′). We highlight the benefits of this approach in terms of value function transfer, learning within redundant action spaces, and learning off-policy from state observations generated by sub-optimal or completely random policies.
Open-ended search algorithms are relatively unstudied in the field of AI safety and yet are starting to show increasing promise as a path to producing advanced AI. This paper discusses the safety implications of open-ended search algorithms in AI.
Introduces Go-Explore, an exploration algorithm capable of solving the grand-challenge hard-exploration Atari games Montezuma′s Revenge and Pitfall.