Research map / Reinforcement learning / Visual and representation RL
Bigger, Better, Faster: Human-level Atari with human-level efficiency
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 BBF instead of SR-SPR | Atari 100K value-based agent · Atari 100K (26 games) | IQM: +0.45 IQM | Helped |
| Increase convolutional-layer reset perturbation from 20% to 50% (ablation) | SR-SPR-based BBF · Atari 100K | performance | Helped |
| Scale Impala-CNN width to 4× (ablation) | BBF · Atari 100K | performance | Helped |
| Scale the standard CNN architecture (ablation) | SR-SPR · Atari 100K | performance | Hurt |
| Use an exponentially decreasing update horizon from 10 to 3 (ablation) | BBF · Atari 100K | performance | Helped |
| Increase the discount factor from 0.97 to 0.997 during training (ablation) | BBF · Atari 100K | performance | Helped |
| Add weight decay of 0.1 using AdamW (ablation) | BBF · Atari 100K | performance | Helped |
| Remove NoisyNets (ablation) | BBF · Atari 100K | task performance | No clear effect |
| Remove the SPR self-supervised objective (ablation) | BBF · Atari 100K | performance | Hurt |
| Use an exponential moving average target network (ablation) | BBF · Atari 100K | performance | Helped |
| Use periodic network resets (ablation) | BBF · Atari 100K | performance | Helped |
| Increase replay ratio (ablation) | BBF · Atari 100K | IQM | Helped |
Closest papers by meaning
- Human-level Atari 200x faster (ICLR 2023 · similarity 0.90)
- Mastering Atari Games with Limited Data (NeurIPS 2021 · similarity 0.89)
- Scaling Offline Model-Based RL via Jointly-Optimized World-Action Model Pretraining (ICLR 2025 · similarity 0.84)
- Uni-RLHF: Universal Platform and Benchmark Suite for Reinforcement Learning with Diverse Human Feedback (ICLR 2024 · similarity 0.84)
- Learning to Play Atari in a World of Tokens (ICML 2024 · similarity 0.82)
- Leveraging Sub-Optimal Data for Human-in-the-Loop Reinforcement Learning (ICLR 2025 · similarity 0.82)
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.