Research map / Reinforcement learning / Visual and representation RL
Synthetic 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 |
|---|---|---|---|
| replace VAE/GAN generator with diffusion | SynthER · halfcheetah-medium-replay | TD3+BC return: 27.1 -> 45.9 | Helped |
| replace offline data with SynthER synthetic data | TD3+BC · D4RL locomotion average | normalized return: 59.0 -> 60.0 | No clear effect |
| replace offline data with SynthER synthetic data | EDAC · D4RL maze2d average | normalized return: 82.6 -> 102.9 | Helped |
| use larger network with SynthER data | TD3+BC · D4RL locomotion average | normalized return: 59.5 -> 65.9 | Helped |
| use larger network with original data | TD3+BC · D4RL locomotion average | normalized return: 59.0 -> 59.5 | No clear effect |
| use larger network | IQL/EDAC · D4RL locomotion | normalized return | No clear effect |
| add SynthER latent data | DrQ+BC · V-D4RL average | normalized return: 51.7 -> 56.5 | Helped |
| add SynthER latent data | BC · V-D4RL average | normalized return: 49.8 -> 53.2 | Helped |
| use only actor features as latent (ablation) | SynthER DrQ+BC · V-D4RL cheetah expert | return: 52.3 -> 43.5 | Hurt |
| use only critic features as latent (ablation) | SynthER DrQ+BC · V-D4RL cheetah expert | return: 52.3 -> 16.0 | Hurt |
| increase denoiser depth from 2 to 6 (residual MLP) (ablation) | SynthER · hopper medium-expert | TD3+BC return: 78.5 -> 101.1 | Helped |
| mix 50-50 real and synthetic data (ablation) | SynthER TD3+BC · D4RL locomotion average | normalized return: 60.0 -> 59.2 | No clear effect |
Closest papers by meaning
- Prioritized Generative Replay (ICLR 2025 · similarity 0.90)
- Actor Prioritized Experience Replay (Abstract Reprint) (AAAI 2024 · similarity 0.89)
- Attentive Experience Replay (AAAI 2020 · similarity 0.89)
- Revisiting Fundamentals of Experience Replay (ICML 2020 · similarity 0.89)
- Remember and Forget for Experience Replay (ICML 2019 · similarity 0.88)
- Prioritized Level Replay (ICML 2021 · 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.