Research map / Reinforcement learning / MDPs and planning
Policy Teaching via Environment Poisoning: Training-time Adversarial Attacks against Reinforcement Learning
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 reward poisoning that preserves rewards for target-policy actions | Online UCRL learner · 4-state MDP | average mismatch | Helped |
| Use dynamics poisoning that preserves transitions for target-policy actions | Online UCRL learner · 4-state MDP | average mismatch | Helped |
| Preserve feedback on target-policy actions during online poisoning (ablation) | Online reward and dynamics attacks · 4-state MDP | average attack cost | Helped |
| Allow poisoning to alter target-policy actions (ablation) | Online reward and dynamics attacks · 4-state MDP | average attack cost | Hurt |
| Increase the robustness margin epsilon | Offline reward poisoning · 4-state MDP | attack cost | Hurt |
| Increase the robustness margin epsilon | Offline dynamics poisoning · 4-state MDP | feasibility: epsilon > 0.85 | Hurt |
| Use unconstrained generic reward or dynamics poisoning instead of non-target-only poisoning (ablation) | Offline poisoning attacks · 4-state MDP | attack cost | Helped |
| Scale the MDP from 4 states to 100 states | Reward poisoning optimization · 4-state and 100-state MDPs | runtime (s): 0.0110 -> 0.8326 | Hurt |
| Scale the MDP from 4 states to 100 states | Dynamics poisoning optimization · 4-state and 100-state MDPs | runtime (s): 3.3010 -> 157.2484 | Hurt |
Closest papers by meaning
- Optimal Attack and Defense for Reinforcement Learning (AAAI 2024 · similarity 0.91)
- Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL (ICLR 2024 · similarity 0.89)
- Adversarial Inception Backdoor Attacks against Reinforcement Learning (ICML 2025 · similarity 0.89)
- Multi-level Certified Defense Against Poisoning Attacks in Offline Reinforcement Learning (ICLR 2025 · similarity 0.89)
- PolicyGuard: Towards Test-time and Step-level Adversary Defense for Reinforcement Learning Agent (ICML 2026 · similarity 0.88)
- Stealthy Imitation: Reward-guided Environment-free Policy Stealing (ICML 2024 · similarity 0.88)
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.