Research map / Reinforcement learning / Visual and representation RL
Generalization in Reinforcement Learning with Selective Noise Injection and Information Bottleneck
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 the Information Bottleneck Actor Critic (IBAC) | PPO · Multiroom | learning performance | Helped |
| Add Selective Noise Injection (SNI) to IBAC | IBAC · Multiroom, 2- and 3-room layouts | success rate: 43% and 21% | Helped |
| Add SNI to IBAC (ablation) | IBAC · Coinrun | test performance | Helped |
| Use SNI with lambda=0.5 rather than lambda=1 (ablation) | IBAC-SNI · Coinrun | test performance | Helped |
| Use SNI with lambda=1 rather than lambda=0.5 (ablation) | Dropout-SNI · Coinrun | learning speed | Helped |
| Add Dropout to weight decay and data augmentation (ablation) | PPO · Coinrun | test performance | Hurt |
| Add Batch Normalization to weight decay and data augmentation (ablation) | PPO · Coinrun | test performance | Hurt |
| Use deterministic rather than stochastic policy updates only (lambda=1) (ablation) | Dropout-SNI · Coinrun | test performance | Helped |
| Add IBAC without SNI (ablation) | PPO · Coinrun | learning speed | Helped |
| Train IBAC longer without additional regularization (ablation) | IBAC · Coinrun | test performance | Hurt |
| Use SNI with the deterministic policy only (lambda=1) (ablation) | SNI · Multiroom, 1-room layout | performance | Hurt |
| Use SNI to reduce policy stochasticity (ablation) | PPO · Multiroom, more complex layouts | performance | Helped |
Closest papers by meaning
- Regularization Matters in Policy Optimization - An Empirical Study on Continuous Control (ICLR 2021 · similarity 0.90)
- On the Power of Pre-training for Generalization in RL: Provable Benefits and Hardness (ICML 2023 · similarity 0.90)
- Regularized Parameter Uncertainty for Improving Generalization in Reinforcement Learning (CVPR 2024 · similarity 0.90)
- A Connection between One-Step RL and Critic Regularization in Reinforcement Learning (ICML 2023 · similarity 0.90)
- Learning to Trust Bellman Updates: Selective State-Adaptive Regularization for Offline RL (ICML 2025 · similarity 0.89)
- Doubly Mild Generalization for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
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.