Research map / Reinforcement learning / Visual and representation RL
Never Give Up: Learning Directed Exploration Strategies
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 a family of 32 exploration/exploitation mixtures with shared weights | NGU · Atari hard-exploration games | score | Helped |
| Add RND as a long-term intrinsic-reward modulator (ablation) | NGU(N=32) · Atari hard-exploration games | score | Helped |
| Add RND as a long-term intrinsic-reward modulator (ablation) | NGU(N=32) · Pitfall! | data efficiency | Hurt |
| Increase cross-mixture replay data ratio to 0.5 (ablation) | NGU(N=32) · Atari hard-exploration games | average score | Hurt |
| Increase the number of mixtures (ablation) | NGU · Atari hard-exploration games | average score | Helped |
| Set maximum intrinsic reward weight to 0.2 instead of 0.3 (ablation) | NGU(N=32) · Atari hard-exploration games | average score | Hurt |
| Set maximum intrinsic reward weight to 0.5 instead of 0.3 (ablation) | NGU(N=32) · Atari hard-exploration games | average score | Hurt |
| Remove extrinsic reward (ablation) | NGU(N=32) · Atari dense-reward games | score | Hurt |
| Use the NGU(N=32) exploitative policy rather than the single-mixture policy (ablation) | NGU · Pong | score: -9.4±2.6 -> 19.6±0.1 | Helped |
| Use the NGU(N=32) method rather than R2D2 | NGU · Breakout | score: 837.7 -> 532.8±16.5 | Hurt |
| Use the NGU(N=32) method rather than R2D2 | NGU · Beam Rider | score: 188.2k -> 68.7k±11.1k | Hurt |
| Use a learned controllable-state embedding rather than a fixed random projection (ablation) | NGU exploratory policy · Random Disco Maze | percentage of unique positions visited | Helped |
Closest papers by meaning
- Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration (NeurIPS 2026 · similarity 0.89)
- Improving Exploration in Evolution Strategies for Deep Reinforcement Learning via a Population of Novelty-Seeking Agents (NeurIPS 2018 · similarity 0.89)
- Redeeming intrinsic rewards via constrained optimization (NeurIPS 2022 · similarity 0.89)
- Successor-Predecessor Intrinsic Exploration (NeurIPS 2023 · similarity 0.88)
- Individual Contributions as Intrinsic Exploration Scaffolds for Multi-agent Reinforcement Learning (ICML 2024 · similarity 0.88)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · 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.