Research map / Adversarial robustness and security / Black-box and transfer attacks
Provably Efficient Black-Box Action Poisoning Attacks Against Reinforcement Learning
Guanlin Liu, Lifeng Lai
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 |
|---|---|---|---|
| Use the α-portion attack, replacing non-target actions with the target action or a worst action under the target policy | Any RL agent with sublinear regret in a tabular episodic MDP · Tabular episodic MDP, white-box | expected cost and loss: E[Cost(K, H)] ≤ E[Loss(K, H)] ≤ R(T)/(α Δ_min) | Improved |
| Estimate target-policy Q-values with importance sampling and explore low-value actions using lower confidence bounds | Efficient RL agents with sublinear expected dynamic regret · Tabular episodic MDP, black-box | cost and loss: Cost(K, H) ≤ Loss(K, H) ≤ H(R(T) + 2H^2√(log(1/p)R(T)))/Δ_min + 307SAH^4log(2SAT/p)/Δ_min^2 | Improved |
| Use LCB-H instead of the white-box 1/H-portion attack | UCB-H, UCB-B, and UCBVI-CH · Periodic 1-d grid world | cumulative attack cost and loss | No clear effect |
| Use LCB-H instead of the white-box 1/H-portion attack | UCB-H, UCB-B, and UCBVI-CH · 4-by-4 grid world | cumulative attack cost and loss | No clear effect |
Closest papers by meaning
- Optimal Attack and Defense for Reinforcement Learning (AAAI 2024 · similarity 0.87)
- Multi-level Certified Defense Against Poisoning Attacks in Offline Reinforcement Learning (ICLR 2025 · similarity 0.86)
- BadRL: Sparse Targeted Backdoor Attack against Reinforcement Learning (AAAI 2024 · similarity 0.86)
- Beyond Worst-case Attacks: Robust RL with Adaptive Defense via Non-dominated Policies (ICLR 2024 · similarity 0.86)
- Adversarial Inception Backdoor Attacks against Reinforcement Learning (ICML 2025 · similarity 0.86)
- Is Poisoning a Real Threat to DPO? Maybe More So Than You Think (AAAI 2025 · similarity 0.86)
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.
