Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Distributed Prioritized 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 |
|---|---|---|---|
| Scale the number of actors from 8 to 256 | Ape-X DQN · Atari (six-game subset) | performance | Helped |
| Increase replay capacity | Ape-X DQN · Atari | performance | Helped |
| Use the full range of actor exploration rates instead of a fixed set of six values (ablation) | Ape-X DQN · Atari | performance | Helped |
| Duplicate each transition eight times with 32 actors instead of gathering distinct transitions with 256 actors (ablation) | Ape-X DQN · Atari | performance | Hurt |
| Use a smaller replay capacity (ablation) | Ape-X DQN · Atari | performance | Hurt |
| Increase batch size up to 512 when using many actors (ablation) | Ape-X DQN · Atari | performance | Helped |
| Increase learning rate to 0.00025 with larger batch sizes (ablation) | Ape-X DQN · Atari | training stability | Hurt |
| Lower learning rate to 0.00025 / 8 (ablation) | Ape-X DQN · Atari | performance | No clear effect |
| Vary the prioritization exponent within [0.3, 0.7] (ablation) | Ape-X DPG · continuous control | performance | No clear effect |
| Increase prioritization exponent beyond 0.7 (ablation) | Ape-X DPG · continuous control | training stability | Hurt |
| Add more actors while keeping the learner update rate unchanged | Ape-X DQN · Atari | performance | Helped |
Closest papers by meaning
- Actor Prioritized Experience Replay (Abstract Reprint) (AAAI 2024 · similarity 0.89)
- Recurrent Experience Replay in Distributed Reinforcement Learning (ICLR 2019 · similarity 0.89)
- Off-Policy Actor-Critic with Shared Experience Replay (ICML 2020 · similarity 0.88)
- Synthetic Experience Replay (NeurIPS 2023 · similarity 0.88)
- Prioritized Generative Replay (ICLR 2025 · similarity 0.87)
- Cleanba: A Reproducible and Efficient Distributed Reinforcement Learning Platform (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.