Bigger, Better, Faster: Human-level Atari with human-level efficiency
This work addresses sample-efficient reinforcement learning for the Atari domain, representing an incremental improvement with specific gains in benchmark performance.
The paper tackles the problem of achieving super-human performance in the Atari 100K benchmark with a value-based RL agent called BBF, which scales neural networks for value estimation and incorporates design choices for sample efficiency, resulting in human-level efficiency and performance.
We introduce a value-based RL agent, which we call BBF, that achieves super-human performance in the Atari 100K benchmark. BBF relies on scaling the neural networks used for value estimation, as well as a number of other design choices that enable this scaling in a sample-efficient manner. We conduct extensive analyses of these design choices and provide insights for future work. We end with a discussion about updating the goalposts for sample-efficient RL research on the ALE. We make our code and data publicly available at https://github.com/google-research/google-research/tree/master/bigger_better_faster.