Research map / Reinforcement learning / MDPs and planning
Reinforcement Learning with a Terminator
Guy Tennenholtz, Nadav Merlis, Lior Shani, Shie Mannor, Uri Shalit, Gal Chechik, Assaf Hallak, Gal Dalal
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| remove optimism (use ensemble mean) (ablation) | TermPG · MinAtar Breakout | mean total reward: 8.2 -> 1.3 | Hurt |
| remove optimism (use ensemble mean) (ablation) | TermPG · MinAtar Space Invaders | mean total reward: 9.7 -> 5.2 | Hurt |
| remove optimism (use ensemble mean) (ablation) | TermPG · Backseat Driver coin avoidance | mean total reward: 8.7 -> 7.8 | Mixed |
| remove dynamic discount factor (gamma 0.99) (ablation) | TermPG · MinAtar Breakout | mean total reward: 8.2 -> 0.5 | Hurt |
| remove dynamic discount factor (gamma 0.99) (ablation) | TermPG · Backseat Driver human termination | mean total reward: 8.3 -> 5.9 | Hurt |
| remove dynamic discount factor (gamma 0.99) (ablation) | TermPG · MinAtar Space Invaders | mean total reward: 9.7 -> 4.4 | Hurt |
| replace min-ensemble optimism with mean minus alpha*std (ablation) | TermPG · MinAtar Space Invaders | mean total reward: 9.7 -> 2.8 | Hurt |
| replace min-ensemble optimism with mean minus alpha*std (ablation) | TermPG · MinAtar Breakout | mean total reward: 8.2 -> 10.9 | Improved |
| replace min-ensemble optimism with mean minus alpha*std (ablation) | TermPG · Backseat Driver coin avoidance | mean total reward: 8.7 -> 7.6 | Hurt |
| window misspecification x0.5 (ablation) | TermPG · MinAtar Space Invaders | mean total reward: 9.7 -> 9.7 | No clear effect |
| window misspecification x2 (ablation) | TermPG · MinAtar Breakout | mean total reward: 8.2 -> 10.3 | Mixed |
| add termination reward-shaping penalty to TermPG | TermPG · MinAtar Breakout | mean total reward: 8.2 -> 5.1 | Hurt |
Closest papers by meaning
- Horizon-Free and Variance-Dependent Reinforcement Learning for Latent Markov Decision Processes (ICML 2023 · similarity 0.87)
- Learning With Options That Terminate Off-Policy (AAAI 2018 · similarity 0.86)
- Horizon-Free Regret for Linear Markov Decision Processes (ICLR 2024 · similarity 0.86)
- RL in Latent MDPs is Tractable: Online Guarantees via Off-Policy Evaluation (NeurIPS 2024 · similarity 0.85)
- DARLING: Detection Augmented Reinforcement Learning with Non-Stationary Guarantees (NeurIPS 2026 · similarity 0.85)
- Model-Based Exploration in Monitored Markov Decision Processes (ICML 2025 · similarity 0.85)
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.
