Research map / Reinforcement learning / Policy gradient and actor-critic
Sample Efficient Deep Reinforcement Learning via Uncertainty Estimation
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 |
|---|---|---|---|
| Apply IV-RL weighting and variance ensembles | DQN · LunarLander | episodes to solved score: 305 -> 227 episodes (BootstrapDQN median to IV-DQN median) | Helped |
| Apply IV-RL weighting and variance ensembles | DQN · Cartpole-Noise | episodes to solved score: 165 -> 112 episodes (SunriseDQN median to IV-DQN median) | Helped |
| Apply IV-RL weighting and variance ensembles | DQN · MountainCar | episodes to solved score: 149 -> 163 episodes (BootstrapDQN median to IV-DQN median) | No clear effect |
| Apply IV-RL weighting and variance ensembles | SAC · Walker, HalfCheetah, Hopper, Ant, SlimHuman. | average return: 100k steps: IV-SAC leads all five; 200k steps: IV-SAC leads four of five, while SunriseSAC leads Ant | Mixed |
| Add BIV weighting to var-ensemble DQN (ablation) | IV-DQN · LunarLander and Cartpole-Noise | sample efficiency | Helped |
| Use variance ensembles instead of sampled ensembles (ablation) | IV-DQN · LunarLander and Cartpole-Noise | sample efficiency | Helped |
| Use a single variance network instead of a variance ensemble (ablation) | IV-DQN · LunarLander and Cartpole-Noise | sample efficiency | Hurt |
| Remove BIV weighting from the var-ensemble method (ablation) | IV-SAC · Walker and HalfCheetah | performance | Hurt |
| Use sampled ensemble variance instead of variance-ensemble estimates (ablation) | IV-SAC · continuous control environments | performance | Hurt |
| Replace BIV weights with SUNRISE weights (ablation) | IV-SAC · continuous control environments | performance | Hurt |
| Replace BIV weights with UWAC inverse-variance weights (ablation) | IV-SAC · continuous control environments | performance | Hurt |
Closest papers by meaning
- Deep Reinforcement Learning in a Handful of Trials using Probabilistic Dynamics Models (NeurIPS 2018 · similarity 0.91)
- Uncertainty-Aware Reward-Free Exploration with General Function Approximation (ICML 2024 · similarity 0.88)
- SUMO: Search-Based Uncertainty Estimation for Model-Based Offline Reinforcement Learning (AAAI 2025 · similarity 0.88)
- Dynamic Uncertainty Estimation for Offline Reinforcement Learning (AAAI 2025 · similarity 0.88)
- Adversarial Diffusion for Robust Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Sample Complexity of Distributionally Robust Average-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.