Research map / Reinforcement learning / Policy gradient and actor-critic
Prioritizing Samples in Reinforcement Learning with Reducible Loss
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 PER priorities with reducible-loss priorities | SAC · DMC Quadruped Run | return: 496.42 $\pm$ 216.01 -> 833.92 $\pm$ 81.05 | Helped |
| Replace PER priorities with reducible-loss priorities | SAC · Gym HalfCheetah-v2 | return: 9549.42 $\pm$ 917.92 -> 11590.63 $\pm$ 670.36 | Helped |
| Replace PER priorities with reducible-loss priorities | DQN · MinAtar Seaquest | score: 6.02 $\pm$ 1.92 -> 18.13 $\pm$ 1.25 | Helped |
| Replace PER priorities with reducible-loss priorities | DQN · MinAtar Space Invaders | score: 37.36 $\pm$ 4.45 -> 38.54 $\pm$ 2.60 | Mixed |
| Replace PER priorities with reducible-loss priorities | DQN · Forgetting Task | success rate: Task A: 0.29 (0.19, 0.39) -> 0.63 (0.52, 0.74); Task B: 0.26 (0.16, 0.36) -> 0.74 (0.65, 0.84) | Helped |
| Replace PER priorities with reducible-loss priorities | SAC · Stochastic DMC Quadruped Run | validation TD error: 5.22 (3.39, 7.05) -> 0.19 (0.17, 0.2) | Helped |
| Use exponential-linear mapping instead of clipping negative ReLo values to zero (ablation) | ReLo · DMC subset | performance | Hurt |
| Prioritize by TD loss rather than reducible loss | SAC · Gym environments | return: Baseline 2795.18 $\pm$ 659.19; PER 2175.09 $\pm$ 456.11 (Gym Hopper) | Hurt |
Closest papers by meaning
- An Equivalence between Loss Functions and Non-Uniform Sampling in Experience Replay (NeurIPS 2020 · similarity 0.91)
- Transfer of Samples in Policy Search via Multiple Importance Sampling (ICML 2019 · similarity 0.90)
- Prioritized Generative Replay (ICLR 2025 · similarity 0.89)
- Sample-Efficiency in Multi-Batch Reinforcement Learning: The Need for Dimension-Dependent Adaptivity (ICLR 2024 · similarity 0.86)
- Demystifying the Paradox of Importance Sampling with an Estimated History-Dependent Behavior Policy in Off-Policy Evaluation (ICML 2025 · similarity 0.86)
- The Courage to Stop: Overcoming Sunk Cost Fallacy in Deep Reinforcement Learning (ICML 2025 · 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.