MicroPatch: Directed Backdoor Erasing via Victim Parameter Decoupling

Summary

MicroPatch is a backdoor erasing approach for deep neural networks that targets the model parameters influenced by backdoor triggers. It reconstructs trigger patterns through reverse engineering, uses influence functions to separate victim parameter components from clean ones, and patches those components. The authors report that across four datasets and four representative backdoor attacks, plus spatial-frequency and frequency-domain attacks, MicroPatch reduces attack success rates more than existing methods while keeping high classification accuracy.

Mitigation

The source describes MicroPatch itself as the proposed mitigation: patching decoupled victim parameter components to purify the model. No separate fix, patched version or configuration change is stated.

Topics

• Adversarial machine learning

Related items

• InfoExploring Backdoor Vulnerabilities and Fairness Challenges in Generative Models: A SurveySimilar attack · ACM Digital Library (TOPS, DTRAP, CSUR)
• MediumCVE-2026-108940: Mistral.rs path traversal through calibration_apply save_cimatrix argumentSimilar attack · NVD/CVE Database
• MediumCVE-2026-108939: mistral.rs server-side request forgery in is_global_ip validationSimilar attack · NVD/CVE Database
• HighCVE-2026-84317: Nginx UI MCP config add tool skips validation, allowing dangerous directivesSimilar attack · NVD/CVE Database
• InfoAnthropic takes risky AI tests offline after models breach govt sitesSimilar attack · BleepingComputer