Research map / Reinforcement learning / Visual and representation RL
Beyond Noisy-TVs: Noise-Robust Exploration Via Learning Progress Monitoring
Zhibo Hou, Zhiyu An, Wan Du
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 |
|---|---|---|---|
| reward learning progress with an error model instead of prediction error or similarity | LPM · Noisy MNIST | steps until intrinsic reward converges to zero: about 400 (AMA) -> about 150 (LPM) | Improved |
| use LPM instead of similarity-based EDT bonus | LPM · Noisy MNIST | intrinsic reward on stochastic vs deterministic transitions | Improved |
| use LPM intrinsic reward for exploration with noisy visual input | LPM · MiniWorld 3D maze, average over three conditions | visited states: +95.3 states over second best (7.6%) | Improved |
| add action noise (random CIFAR image on idle action) to MiniWorld | baseline intrinsic-motivation methods · MiniWorld 3D maze | state coverage | Hurt |
| LPM in deterministic MountainCar versus Ensemble | LPM · MountainCar-Continuous deterministic | state coverage (%): 91.22 -> 76.50 | Hurt |
| use LPM under stochastic observation noise in MountainCar | LPM · MountainCar-Continuous stochastic | state coverage (%): 76.50 -> 67.04 (12.4% drop vs 33.1% to 85.9% for baselines) | Improved |
| use LPM under action noise on Atari | LPM · Atari with idle-action noise | extrinsic reward: 3.9% drop in Space Invader, 4.7% drop in UpNDown, 0.3% improvement in Ms PacMan | Improved |
| use LPM on clean and noisy Atari versus strong baselines | LPM · Atari (6 settings) | extrinsic reward: best in 4 of 6 | Mixed |
| use LPM on Montezuma's Revenge | LPM · Montezuma's Revenge | steps to non-trivial score: RND 50M steps -> LPM 20M steps | Improved |
| use pointwise error difference instead of expected previous error from the error model (ablation) | LPM · theory (Gaussian likelihood) | monotonicity with information gain | Hurt |
| LPM on MiniGrid Lava Crossing versus ICM | LPM · MiniGrid Lava Crossing | PPO success rate | No clear effect |
Closest papers by meaning
- CIG: Exploration via Conditional Information Gain (NeurIPS 2026 · similarity 0.88)
- Beyond Optimism: Exploration With Partially Observable Rewards (NeurIPS 2024 · similarity 0.88)
- Rethinking Exploration in Reinforcement Learning with Effective Metric-Based Exploration Bonus (NeurIPS 2024 · similarity 0.87)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration (NeurIPS 2026 · similarity 0.87)
- Improving Intrinsic Exploration by Creating Stationary Objectives (ICLR 2024 · 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.
