Research map / Reinforcement learning / Policy gradient and actor-critic
Safe Policy Improvement with Baseline Bootstrapping
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 |
|---|---|---|---|
| Bootstrap insufficiently sampled state-action pairs with the baseline policy | SPIBB · 5×5 gridworld with baseline-collected data | 1%-CVaR performance | Helped |
| Use Π_≤b-SPIBB instead of Π_b-SPIBB, allowing the policy to reduce probability on poorly performing bootstrapped actions (ablation) | SPIBB · 5×5 gridworld with baseline-collected data | mean performance | Helped |
| Use Π_≤b-SPIBB instead of Π_b-SPIBB (ablation) | SPIBB · 5×5 gridworld with baseline-collected data | 1%-CVaR performance | Helped |
| Increase the bootstrapping threshold N_wedge (ablation) | SPIBB · 5×5 gridworld | 1%-CVaR performance | Mixed |
| Replace baseline-collected data with uniformly collected data (ablation) | SPIBB · 5×5 gridworld | 1%-CVaR performance | Mixed |
| Use Π_≤b-SPIBB rather than RaMDP | Random MDPs with η=0.9 · Random MDPs | 1%-CVaR performance | Helped |
| Increase N_wedge from too-low values (ablation) | Π_≤b-SPIBB · Random MDPs with η=0.9 | 1%-CVaR performance | Helped |
| Set N_wedge=0, removing SPIBB bootstrapping (ablation) | SPIBB-DQN · Helicopter navigation | mean performance | Hurt |
| Use SPIBB-DQN with N_wedge≥1 instead of vanilla DQN (ablation) | SPIBB-DQN · Helicopter navigation | mean performance | Helped |
| Use SPIBB-DQN with N_wedge≥8 instead of vanilla DQN (ablation) | SPIBB-DQN · Helicopter navigation | 10%-CVaR performance | Helped |
| Attenuate transition noise (ablation) | SPIBB-DQN · Helicopter navigation | mean performance | Hurt |
| Increase transition noise (ablation) | SPIBB-DQN · Helicopter navigation | safe policy improvement | Mixed |
Closest papers by meaning
- Scalable Safe Policy Improvement via Monte Carlo Tree Search (ICML 2023 · similarity 0.89)
- Multi-Objective SPIBB: Seldonian Offline Policy Improvement with Safety Constraints in Finite MDPs (NeurIPS 2021 · similarity 0.89)
- Safe Policy Improvement with Baseline Bootstrapping in Factored Environments (AAAI 2019 · similarity 0.89)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.86)
- A Provable Approach for End-to-End Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- Offline Safe Reinforcement Learning Using Trajectory Classification (AAAI 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.