AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

BIRD: Generalizable Backdoor Detection and Removal for Deep Reinforcement Learning

Xuan Chen, Wenbo Guo, Guanhong Tao, Xiangyu Zhang, Dawn Song

NeurIPS 2023 · Adversarial robustness and security › Backdoor and poisoning · approach: Detecting poisoned samples · 4 citations (OpenAlex)

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 changedWhereMeasuredEffect
Replace reward-based detection metric with ||∆||0 (ablation)BIRD detection · QbertF1 score: 0.909 -> 0.571Hurt
Replace reward-based detection metric with ||∆||0 (ablation)BIRD detection · COMAF1 score: 1.000 -> 0.347Hurt
Replace reward-based detection metric with ||∆||0 (ablation)BIRD detection · YSNPF1 score: 0.667 -> 0.263Hurt
Add neuron re-initialization before retraining (ablation)BIRD removal · Qbert poisonedreward: 0.0%±0.0% -> 12694±2194Improved
Add neuron re-initialization before retraining (ablation)BIRD removal · COMA poisonedwinning rate: 6.7%±2.1% -> 90.7%±3.0%Improved
Add neuron re-initialization before retraining (ablation)BIRD removal · YSNP poisonedwinning rate: 29.5%±4.1% -> 46.7%±3.9%Improved
Use BIRD instead of Finetuningbackdoor removal · SpaceInvaders poisonedreward: 21±65 -> 543±52Improved
Use BIRD instead of PDbackdoor removal · SpaceInvaders cleanreward: 238±94 -> 529±92Improved
Use BIRD instead of PDbackdoor removal · Qbert poisonedreward: 6180±1817 -> 12694±2194Improved
Apply BIRD in clean environment versus original policyclean performance · SpaceInvadersreward: 610±97 -> 529±92Mixed
Change trigger size to 5 × 5BIRD detection · SpaceInvadersF1 score: 1.0 -> 0.833Hurt
Change trigger shape to circleBIRD detection · SpaceInvadersF1 score: 1.0 -> 0.909Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy