Research map / Reinforcement learning / Visual and representation RL
Curiosity-Bottleneck: Exploration By Distilling Task-Specific Novelty
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 KL (compression) term to value-predictive objective, i.e. CB vs CB-noKL (ablation) | CB intrinsic reward · MNIST and Fashion-MNIST static classification with distractions | SDR score | Helped |
| Use CB intrinsic reward instead of task-agnostic baselines (RND, SimHash) | exploration bonus on static image classification · Fashion-MNIST with Box, Noise and Object distractions | SDR score | Helped |
| Use CB intrinsic reward instead of RND/SimHash in Treasure Hunt | PPO agent · Treasure Hunt grid world with Random Box distraction | mean episodic reward | Helped |
| Use CB intrinsic reward instead of baselines in distractive Atari games | PPO agent (CNN policy) · Gravitar, Montezuma's Revenge, Solaris with Random Box distraction | mean episodic return | Helped |
| Use recurrent (RNN) policy instead of CNN policy | CB and baselines · Atari distractive games | mean episodic return | No clear effect |
| Use CB in environments without distractions | CB intrinsic reward · Atari hard exploration games without distraction | mean episodic return | Mixed |
Closest papers by meaning
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Catching Two Birds with One Stone: Reward Shaping with Dual Random Networks for Balancing Exploration and Exploitation (ICML 2025 · similarity 0.88)
- MaxInfoRL: Boosting exploration in reinforcement learning through information gain maximization (ICLR 2025 · similarity 0.87)
- Rethinking Exploration in Reinforcement Learning with Effective Metric-Based Exploration Bonus (NeurIPS 2024 · similarity 0.87)
- Wonder Wins Ways: Curiosity-Driven Exploration through Multi-Agent Contextual Calibration (NeurIPS 2025 · similarity 0.87)
- CIG: Exploration via Conditional Information Gain (NeurIPS 2026 · 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.