Research map / Reinforcement learning / Visual and representation RL
Agent57: Outperforming the Atari Human Benchmark
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Agent57 instead of prior Atari agents | Atari 57 games · Atari 57 games | number of games > human: 52 -> 57 | Helped |
| Add a meta-controller to choose policies adaptively (ablation) | R2D2 (Retrace) · Atari 57 games | capped mean HNS: 94.20 -> 96.93 | Helped |
| Add separate intrinsic and extrinsic value networks (ablation) | Agent57 · Atari challenging set | capped mean HNS: greater than 20% | Helped |
| Add a meta-controller (ablation) | R2D2 · Atari challenging set | capped mean HNS: close to 20% CHNS | Helped |
| Increase BPTT window from 80 to 160 (ablation) | R2D2 and Agent57 · Atari challenging set | capped mean HNS | Helped |
| Increase BPTT window from 80 to 160 (ablation) | R2D2 and Agent57 · Solaris | score | Helped |
| Use a fixed discount factor of 0.9999 across games (ablation) | R2D2 · Atari challenging set | end performance | Hurt |
Closest papers by meaning
- Human-level Atari 200x faster (ICLR 2023 · similarity 0.94)
- Multi-Game Decision Transformers (NeurIPS 2022 · similarity 0.90)
- Model Based Reinforcement Learning for Atari (ICLR 2020 · similarity 0.90)
- Mastering Atari with Discrete World Models (ICLR 2021 · similarity 0.89)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 2024 · similarity 0.88)
- FightLadder: A Benchmark for Competitive Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.