AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

SRD: Reinforcement-Learned Semantic Perturbation for Backdoor Defense in VLMs

Shuhan Xu, Siyuan Liang, Hongling Zheng, Aishan Liu, Xinbiao Wang, Yong Luo, Fu Lin, Leszek Rutkowski, Dacheng Tao

AAAI 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Backdoors in large multimodal models · 11 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
use DQN instead of random red masking as the mask policy (ablation)SRD · COCO, BadNetASR (%): 73.5 -> 45.7Improved
use DQN instead of PPO (ablation)SRD · COCO, BadNetASR (%): 72.0 -> 45.7Improved
use DQN instead of SAC (ablation)SRD · COCO, VLOODASR (%): 0.9 -> 38.5Hurt
random red masking alone, without learned policy (ablation)SRD · COCO, TrojVLMASR (%): 3.8 vs 3.6No clear effect
apply SRD red-mask retraining defensebackdoored OpenFlamingo-3B · COCO, TrojVLMASR (%): 99.8 -> 3.6Improved
apply SRD to BadNet and Blended attacksbackdoored OpenFlamingo-3B · COCOASR (%): 99.9 -> 45.7 (BadNet), 99.5 -> 57.4 (Blended)Improved
apply SRD instead of ABL/VDC/CT on Shadowcastbackdoored OpenFlamingo-3B · COCO, ShadowcastASR (%): CT 6.7 vs SRD 5.6Mixed
apply SRD defense on clean inputsbackdoored OpenFlamingo-3B · COCO clean test setCIDEr (%): 103.9 clean model; SRD range 88.6 to 101.2Hurt
lower the poisoning rate (1% vs 10%) (ablation)SRD · BadNetASRImproved
SRD against Flickr30k BadNetbackdoored OpenFlamingo-3B · Flickr30k, BadNetASR (%): 98.2 -> 55.6Improved

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