Research map / Reinforcement learning / Policy gradient and actor-critic
Multi-step Greedy Reinforcement Learning Algorithms
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 κ-PI-DQN or κ-VI-DQN with tuned κ instead of DQN (κ=1) | DQN · Atari benchmarks | return | Helped |
| Use κ-PI-TRPO or κ-VI-TRPO with tuned κ instead of TRPO (κ=1) | TRPO · MuJoCo benchmarks | return | Helped |
| Set κ to 0 instead of the best κ (ablation) | κ-PI-DQN · Atari benchmarks | return | Hurt |
| Allocate samples across multiple surrogate-problem iterations rather than use Nκ=T (ablation) | κ-PI-DQN and κ-VI-DQN · Atari benchmarks | return | Helped |
| Use Nκ=T rather than the recommended C_FA allocation (ablation) | κ-PI-TRPO and κ-VI-TRPO · Walker, Ant, HalfCheetah | return | Hurt |
| Use Nκ=T rather than the recommended C_FA allocation (ablation) | κ-PI-TRPO and κ-VI-TRPO · HumanoidStand, Swimmer, Hopper | return | No clear effect |
| Set C_FA to 0.05 rather than the tested alternatives (ablation) | κ-PI-DQN and κ-VI-DQN · Breakout | return | Helped |
| Set C_FA to 0.2 rather than the tested alternatives (ablation) | κ-PI-TRPO and κ-VI-TRPO · Walker | return | Helped |
| Lower γ in vanilla DQN or TRPO (ablation) | DQN and TRPO · Breakout, SpaceInvaders, Walker, Ant | return | Mixed |
| Use κ-PI-TRPO with Nκ=T, equivalent to the described GAE-style update (ablation) | κ-PI-TRPO · MuJoCo benchmarks | return | No clear effect |
| Use a hard target network for π(i−1) during policy improvement (ablation) | κ-PI-DQN · DQN experiments | return | No clear effect |
| Use multi-step greedy updates rather than lower γ directly (ablation) | DQN and TRPO · Atari and MuJoCo benchmarks | return | Helped |
Closest papers by meaning
- Multiple-Step Greedy Policies in Approximate and Online Reinforcement Learning (NeurIPS 2018 · similarity 0.91)
- Beyond the One-Step Greedy Approach in Reinforcement Learning (ICML 2018 · similarity 0.89)
- Tight Regret Bounds for Model-Based Reinforcement Learning with Greedy Policies (NeurIPS 2019 · similarity 0.89)
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Inverting the Bellman Equation: From $Q$-Values to World Models (NeurIPS 2026 · similarity 0.87)
- Multi-Step Likelihood-Ratio Correction for Reinforcement Learning with Verifiable Rewards (NeurIPS 2026 · 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.