AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

BYORn: Bootstrap Your Own Responses to Defend Large Vision-Language Models Against Backdoor Attacks

Ivan Sabolic, Marin Oršić, Josip Šarić, Sven Loncaric

ICML 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Detecting poisoned samples · 0 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
regenerate suspicious responses with EMA model instead of filtering them (BYORn vs BYORn-F) (ablation)BYORn · Blend, p=0.975, LLaVAASR (%): 79.3 -> 1.2Improved
regenerate suspicious responses instead of filtering (ablation)BYORn · Blend, p=0.95, LLaVAASR (%): 57.6 -> 1.1Improved
regenerate responses instead of filtering (ablation)BYORn · Image captioning Flickr, LLaVACIDEr: 42.4 -> 62.0Improved
regenerate responses instead of filtering (ablation)BYORn · Spot the Difference, LLaVACIDEr: 129.8 -> 134.9Improved
add poison-aware minibatching (ablation)BYORn · LLaVA training on LADDepoch time (min): 301 -> 135Improved
add speculative decoding with a LLaVA-OneVision draft modelBYORn · LLaVA training on LADDepoch time (min): 301 -> 294No clear effect
raise percentile threshold p from 0.6 to 0.975 (ablation)BYORn · Blend, LLaVAASR (%): 0.0 -> 1.2No clear effect
lower p to 0.3 (many false positives) (ablation)BYORn · Blend, LLaVACIDEr: 62.2 -> 55.6Hurt
vary poisoning rate from 1% to 50% (ablation)BYORn · VL-Trojan, LLaVA Flickr30kASR (%): 0.0 -> 0.2No clear effect
vary EMA decay lambda from 0.99 to 0.999 (ablation)BYORn · Blend, LLaVACIDEr: 61.8 -> 61.2No clear effect
apply BYORn defense against adaptive image-wide attackLLaVA SFT · Adaptive image-wide, COCOASR (%): 79.4 -> 8.4Improved
apply BYORn against PGD adversarial-perturbation triggerLLaVA SFT · Flickr30kASR (%): 92.0 -> 0.2Improved

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