Summary of Learning to Play Atari in a World Of Tokens, by Pranav Agarwal et al.
Learning to Play Atari in a World of Tokens
by Pranav Agarwal, Sheldon Andrews, Samira Ebrahimi Kahou
First submitted to arxiv on: 3 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach to reinforcement learning is introduced in this paper, focusing on transformer-based models that utilize discrete abstract representations for efficient learning. By incorporating a transformer-decoder for world modeling and a transformer-encoder for learning behavior, the proposed method, DART (Discrete Abstract Representations for Transformers), achieves improved sample efficiency and outperforms previous state-of-the-art methods on the Atari 100k benchmark. The paper also addresses partial observability by aggregating information from past time steps as memory tokens. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Reinforcement learning agents using transformers have become more efficient due to their ability to model extended context, making world models more accurate. However, these methods often rely on continuous representations, which can be limiting for complex tasks like reasoning and planning. To overcome this limitation, the authors introduce DART, a new method that uses discrete abstract representations for both world modeling and learning behavior. By incorporating a transformer-decoder for auto-regressive world modeling and a transformer-encoder for learning behavior, DART achieves better results on the Atari 100k benchmark. |
Keywords
» Artificial intelligence » Decoder » Encoder » Reinforcement learning » Transformer
