Research map / Reinforcement learning / Visual and representation RL
How RL Agents Behave When Their Actions Are Modified
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 Bellman optimality learning rather than black-box reward maximization | RL agents under action modification · Exp Inverting Bandit | expected reward: 0.55 -> -0.46 | Hurt |
| Use black-box reward maximization rather than Bellman optimality learning | RL agents under action modification · Exp Inverting Bandit | expected reward: -0.46 -> 0.55 | Helped |
| Use REINFORCE instead of black-box reward maximization | Exp Inverting Bandit · Exp Inverting Bandit | expected reward: 0.55 -> -0.54 | Hurt |
| Use Empirical Sarsa rather than Q-Learning | RL agents under action modification · Simulation-Oversight | learned policy behavior | Mixed |
| Use Bellman optimality or virtual policy value learning | RL agents under action modification · Off-Switch and Small Whisky-Gold | avoiding harmful modifications | Mixed |
| Use empirical policy value learning | RL agents under action modification · Off-Switch and Small Whisky-Gold | avoiding harmful modifications | Mixed |
| Set scale_sample_evaluations to 1, disabling dynamic uncertainty handling (ablation) | UH-CMA-ES · Simulation-Oversight, Off-Switch, and Small Whisky-Gold | agent performance | No clear effect |
| Increase initial_num_sample_evaluations from 1 to 100 (ablation) | UH-CMA-ES · Simulation-Oversight, Off-Switch, and Small Whisky-Gold | training stability: 1 -> 100 | Helped |
Closest papers by meaning
- Learn to change the world: Multi-level reinforcement learning with model-changing actions (ICML 2026 · similarity 0.90)
- Adaptive Model Design for Markov Decision Process (ICML 2022 · similarity 0.88)
- Explainable Reinforcement Learning via Model Transforms (NeurIPS 2022 · similarity 0.88)
- Reinforcement Learning with Parameterized Actions (AAAI 2016 · similarity 0.87)
- Real-Time Reinforcement Learning (NeurIPS 2019 · similarity 0.87)
- Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis (AAAI 2024 · 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.