Research map / Reinforcement learning / Policy gradient and actor-critic
Sample-Efficient Deep Reinforcement Learning via Episodic Backward Update
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 episodic backward value propagation instead of uniform one-step replay | DQN · 49 Atari games | mean human-normalized score (%): 133.95 -> 253.55 | Helped |
| Use episodic backward value propagation instead of uniform one-step replay | DQN · 49 Atari games | median human-normalized score (%): 40.42 -> 51.55 | Helped |
| Adapt the diffusion factor using parallel learners (ablation) | EBU · 49 Atari games | mean human-normalized score (%): 253.55 -> 275.78 | Helped |
| Adapt the diffusion factor using parallel learners (ablation) | EBU · 49 Atari games | median human-normalized score (%): 51.55 -> 63.80 | Helped |
| Increase training from 10M to 20M frames (ablation) | Adaptive EBU · 49 Atari games | median human-normalized score (%): 63.80 -> 92.50 | Helped |
| Use N-step targets instead of EBU | DQN · MNIST Maze, 20% wall density | mean relative length: 5.44 -> 3.26 | Helped |
| Use EBU instead of N-step targets | DQN · MNIST Maze, 30%-50% wall density | relative length | Mixed |
| Use EBU instead of uniform one-step replay | DQN · MNIST Maze, 20%-50% wall density | relative length | Helped |
| Set the diffusion factor to β=1.0 rather than β=0.5 (ablation) | EBU · Atari games | episode score | Hurt |
| Set the diffusion factor to β=1.0 rather than β=0.5 (ablation) | EBU · MNIST Maze | relative length | Helped |
| Use β=1.0 instead of a diffusion factor below 1 (ablation) | EBU · Atari games | Q-value estimation | Hurt |
| Use episode-level sampling and backward target updates rather than uniform transition sampling | DQN · Atari games | training time | No clear effect |
Closest papers by meaning
- Fast deep reinforcement learning using online adjustments from the past (NeurIPS 2018 · similarity 0.88)
- A State-Distribution Matching Approach to Non-Episodic Reinforcement Learning (ICML 2022 · similarity 0.87)
- Sample-Efficient Reinforcement Learning by Breaking the Replay Ratio Barrier (ICLR 2023 · similarity 0.87)
- Neural Episodic Control (ICML 2017 · similarity 0.86)
- ART for Diffusion Sampling: A Reinforcement Learning Approach to Timestep Scheduling (NeurIPS 2026 · similarity 0.85)
- Symmetric Replay Training: Enhancing Sample Efficiency in Deep Reinforcement Learning for Combinatorial Optimization (ICML 2024 · similarity 0.85)
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.