Research map / Reinforcement learning / Visual and representation RL
Dynamic Bottleneck for Robust Self-Supervised Exploration
Chenjia Bai, Lingxiao Wang, Lei Han, Animesh Garg, Jianye HAO, Peng Liu, Zhaoran Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 DB-bonus exploration (SSE-DB) instead of ICM, Disagreement and Curiosity Bottleneck | PPO with intrinsic reward · 18 Atari games, self-supervised exploration | extrinsic score: best in 15 of 18 tasks | Improved |
| use SSE-DB instead of prediction-error methods such as ICM | PPO with intrinsic reward · Tennis | extrinsic score | Hurt |
| use SSE-DB instead of ICM under random-box noise | PPO with intrinsic reward · Alien and TimePilot with random-box noise | extrinsic score | Improved |
| use SSE-DB instead of ICM under random-box noise | PPO with intrinsic reward · Breakout with random-box noise | extrinsic score | No clear effect |
| use SSE-DB instead of ICM with sticky actions (prob 0.25) | PPO with intrinsic reward · Boxing, Gravitar, MsPacman with sticky actions | extrinsic score | Improved |
| add the compression term I_upper to the DB loss (ablation) | DB model · Atari ablation, with dynamics-irrelevant noise | extrinsic score | Improved |
| add the predictive objective I_pred to the DB loss (ablation) | DB model · Atari ablation | extrinsic score | Improved |
| add the contrastive objective I_nce to the DB loss (ablation) | DB model · Atari ablation | extrinsic score | Improved |
| add the momentum observation encoder (ablation) | DB model · Atari ablation | extrinsic score | Improved |
Closest papers by meaning
- Curiosity-Bottleneck: Exploration By Distilling Task-Specific Novelty (ICML 2019 · similarity 0.91)
- Drop-Bottleneck: Learning Discrete Compressed Representation for Noise-Robust Exploration (ICLR 2021 · similarity 0.89)
- InfoBot: Transfer and Exploration via the Information Bottleneck (ICLR 2019 · similarity 0.87)
- Beyond Noisy-TVs: Noise-Robust Exploration Via Learning Progress Monitoring (ICLR 2026 · similarity 0.87)
- The Variational Bandwidth Bottleneck: Stochastic Evaluation on an Information Budget (ICLR 2020 · similarity 0.87)
- Rethinking Exploration in Reinforcement Learning with Effective Metric-Based Exploration Bonus (NeurIPS 2024 · similarity 0.85)
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.
