Research map / Reinforcement learning / Policy gradient and actor-critic
Safe and Efficient Off-Policy Reinforcement Learning
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 Retrace($\lambda$) traces instead of TB($\lambda$) | experience-replay Q-learning agent · 60 Atari 2600 games | inter-algorithm score: best on 30 games; TB claims 15 of the remainder | Helped |
| use Retrace or TB multi-step returns instead of Q-Learning ($\lambda = 0$) | experience-replay agent · 60 Atari 2600 games | inter-algorithm score | Helped |
| set $\lambda = 1$ rather than smaller values (ablation) | Retrace and TB · 60 Atari 2600 games | inter-algorithm score | Helped |
| vary $\lambda$ for $Q^*(\lambda)$ (sensitivity) (ablation) | $Q^*(\lambda)$ · 60 Atari 2600 games | inter-algorithm score | Hurt |
| use multi-schedule epsilon exploration instead of a single DQN-style schedule (ablation) | multi-threaded agent · Atari 2600 | performance | Helped |
| use Retrace instead of TB across all lambda values | agent · Atari, averaged inter-algorithm score vs lambda | average inter-algorithm score | Helped |
Closest papers by meaning
- Behaviour Policy Optimization: Provably Lower Variance Return Estimates for Off-Policy Reinforcement Learning (AAAI 2026 · similarity 0.90)
- Chaining Value Functions for Off-Policy Learning (AAAI 2022 · similarity 0.89)
- Online and Offline Reinforcement Learning by Planning with a Learned Model (NeurIPS 2021 · similarity 0.89)
- Skill or Luck? Return Decomposition via Advantage Functions (ICLR 2024 · similarity 0.89)
- Asymmetric REINFORCE for off-Policy Reinforcement Learning: Balancing positive and negative rewards (NeurIPS 2025 · similarity 0.88)
- Model Selection for Off-policy Evaluation: New Algorithms and Experimental Protocol (NeurIPS 2025 · 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.