Research map / Reinforcement learning / MDPs and planning
Provable Defense against Backdoor Policies in Reinforcement Learning
Shubham Kumar Bharti, Xuezhou Zhang, Adish Singla, Jerry Zhu
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 |
|---|---|---|---|
| sanitize backdoor policy by projecting states onto safe subspace using n clean samples | backdoor policy (Boxing-Ram, Breakout) · triggered environment | mean empirical return | Improved |
| increase number of clean sanitization samples n (ablation) | sanitized policy · Boxing-Ram and Breakout | mean empirical return | Improved |
| underestimate safe subspace dimension d (ablation) | sanitized policy · Boxing-Ram and Breakout | mean empirical return | Hurt |
| overestimate safe subspace dimension d (ablation) | sanitized policy · Boxing-Ram and Breakout | mean empirical return | Hurt |
| use roughly 2n correlated samples instead of n independent samples (ablation) | sanitization algorithm · Boxing-Ram and Breakout | mean empirical return | No clear effect |
Closest papers by meaning
- PolicyGuard: Towards Test-time and Step-level Adversary Defense for Reinforcement Learning Agent (ICML 2026 · similarity 0.89)
- Optimal Attack and Defense for Reinforcement Learning (AAAI 2024 · similarity 0.87)
- SHINE: Shielding Backdoors in Deep Reinforcement Learning (ICML 2024 · similarity 0.87)
- BadRL: Sparse Targeted Backdoor Attack against Reinforcement Learning (AAAI 2024 · similarity 0.87)
- Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL (ICLR 2024 · similarity 0.86)
- Correlated Proxies: A New Definition and Improved Mitigation for Reward Hacking (ICLR 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.
