Research map / Reinforcement learning / Policy gradient and actor-critic
Large Batch Experience Replay
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 LaBER-mean with up-to-date priorities instead of uniform replay | DQN on MinAtar · MinAtar games | average episode reward | Helped |
| use LaBER instead of DQN with a larger uniform mini-batch | DQN on MinAtar and Atari · MinAtar games and Atari games | average reward | Helped |
| use LaBER with surrogate gradient norms instead of exact gradient norms (ablation) | LaBER-mean · MinAtar games | average episode reward | No clear effect |
| increase the LaBER large-batch multiplier m (ablation) | LaBER-mean with mini-batch size B=32 · MinAtar games | average episode reward | Helped |
| use LaBER-lazy instead of LaBER-mean normalization (ablation) | LaBER · MinAtar games | average episode reward | Mixed |
| use LaBER-max instead of LaBER-mean normalization (ablation) | LaBER · MinAtar games | average episode reward | Hurt |
| add up-to-date LaBER resampling after PER priorities (ablation) | PER · MinAtar games | average episode reward | Helped |
| add up-to-date LaBER resampling after GER priorities (ablation) | GER · MinAtar games | average episode reward | Helped |
| add PER-LaBER or GER-LaBER priority sampling to vanilla LaBER (ablation) | LaBER · MinAtar games | average episode reward | No clear effect |
| use LaBER rather than PER or GER prioritization | DQN, C51, SAC, and TD3 · 5 Atari games and 5 continuous-control tasks | average return | Helped |
| tune PER or GER with alpha=0.6 and c=10^-10 rather than alpha=1 and c=0 (ablation) | PER and GER · MinAtar games | average episode reward | Helped |
Closest papers by meaning
- Actor Prioritized Experience Replay (Abstract Reprint) (AAAI 2024 · similarity 0.91)
- An Equivalence between Loss Functions and Non-Uniform Sampling in Experience Replay (NeurIPS 2020 · similarity 0.91)
- Sample-Efficient Reinforcement Learning by Breaking the Replay Ratio Barrier (ICLR 2023 · similarity 0.90)
- Prioritizing Samples in Reinforcement Learning with Reducible Loss (NeurIPS 2023 · similarity 0.90)
- Prioritized Generative Replay (ICLR 2025 · similarity 0.89)
- When Does Non-Uniform Replay Matter in Reinforcement Learning? (NeurIPS 2026 · 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.