Research map / Reinforcement learning / Policy gradient and actor-critic
Multi-Step Reinforcement Learning: A Unifying Algorithm
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 |
|---|---|---|---|
| Decrease σ by a factor of 0.95 after each episode instead of keeping it fixed | Q(σ) · 19-state random walk | RMS error | Helped |
| Decrease σ by a factor of 0.95 after each episode instead of keeping it fixed | Q(σ) · Stochastic windy gridworld | average return | Helped |
| Use a 3-step backup rather than a 1-step backup (ablation) | Q(σ) · Stochastic windy gridworld | average return | Helped |
| Use an intermediate fixed sampling degree (σ=0.5) rather than either extreme (ablation) | Q(σ) · Stochastic windy gridworld | average return | Helped |
| Decrease σ dynamically rather than keeping it fixed | Q(σ) · Mountain cliff, after 50 episodes | average return per episode: -398.0 -> -406.3 | Mixed |
| Decrease σ dynamically rather than keeping it fixed | Q(σ) · Mountain cliff, after 500 episodes | average return per episode: -167.9 -> -163.7 | Helped |
| Use intermediate σ rather than full sampling (σ=1) (ablation) | Q(σ) · Mountain cliff, after 50 episodes | average return per episode: -447.3 -> -398.0 | Helped |
| Use dynamic σ rather than full sampling (σ=1) (ablation) | Q(σ) · Mountain cliff, after 500 episodes | average return per episode: -173.2 -> -163.7 | Helped |
| Decrease σ over training rather than hold it constant | Q(σ) · 19-state random walk, initial versus asymptotic performance | RMS error | Mixed |
Closest papers by meaning
- Multiple-Step Greedy Policies in Approximate and Online Reinforcement Learning (NeurIPS 2018 · similarity 0.90)
- Beyond the One-Step Greedy Approach in Reinforcement Learning (ICML 2018 · similarity 0.89)
- Long-Horizon Q-Learning: Accurate Value Learning via n-Step Inequalities (NeurIPS 2026 · similarity 0.88)
- Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems (NeurIPS 2025 · similarity 0.88)
- A Unifying View of Linear Function Approximation in Off-Policy RL Through Matrix Splitting and Preconditioning (NeurIPS 2025 · similarity 0.87)
- Efficient Multi-Policy Evaluation for Reinforcement Learning (AAAI 2025 · 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.