Research map / Reinforcement learning / Visual and representation RL
On Bonus Based Exploration Methods In The Arcade Learning Environment
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 |
|---|---|---|---|
| Add a bonus-based exploration method instead of ε-greedy exploration | Rainbow · Montezuma's Revenge | score: 780.0 -> 5199.2 | Helped |
| Use newer bonus methods rather than pseudo-counts | Rainbow · Montezuma's Revenge | score: 5028.2; 5199.2 | No clear effect |
| Use bonus-based exploration rather than ε-greedy exploration | Rainbow · Hard exploration games other than Montezuma's Revenge | score | No clear effect |
| Use ε-greedy rather than bonus-based exploration | Rainbow · Gravitar | score | Helped |
| Add CTS pseudo-count exploration | Rainbow · Full Atari suite | AUC | Mixed |
| Add RND exploration bonus | Rainbow · Easy-exploration Atari games | score | No clear effect |
| Use NoisyNets instead of ε-greedy exploration | Rainbow · Full Atari suite | AUC | Helped |
| Tune bonus-method hyperparameters on the original ALE training set rather than Montezuma's Revenge | CTS and RND · Atari games | score | No clear effect |
| Tune CTS on a different training set | Rainbow + CTS · Seaquest | percentage improvement over ε-greedy: 120% -> 140% | Helped |
| Increase training from 200 million to 1 billion frames | Bonus-based exploration methods · Hard exploration games | score: 200 million; 1 billion | No clear effect |
| Increase training data to 1 billion frames | Rainbow + RND · Montezuma's Revenge | score: 1 billion | Helped |
| Increase training data | Bonus-based exploration methods · Easy-exploration games | score: 1 billion | Helped |
Closest papers by meaning
- Atari-5: Distilling the Arcade Learning Environment down to Five Games (ICML 2023 · similarity 0.89)
- Large-Scale Study of Curiosity-Driven Learning (ICLR 2019 · similarity 0.89)
- Curiosity in Hindsight: Intrinsic Exploration in Stochastic Environments (ICML 2023 · similarity 0.89)
- Rethinking Exploration in Reinforcement Learning with Effective Metric-Based Exploration Bonus (NeurIPS 2024 · similarity 0.88)
- Intelligent Go-Explore: Standing on the Shoulders of Giant Foundation Models (ICLR 2025 · similarity 0.87)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
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.