Research map / Large language models / Instruction and fine-tuning
Enhancing Fine-Tuning Based Backdoor Defense with Sharpness-Aware Minimization
Mingli Zhu, Shaokui Wei, Li Shen, Yanbo Fan, Baoyuan Wu
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 |
|---|---|---|---|
| replace vanilla fine-tuning with sharpness-aware fine-tuning (FT-SAM) | backdoored PreAct-ResNet18 defense · CIFAR-10 5% benign, average over 10 attacks | ASR (%): 38.81 -> 2.47 | Improved |
| replace vanilla fine-tuning with FT-SAM | backdoored PreAct-ResNet18 defense · Tiny ImageNet 5% benign, average over attacks | ASR (%): 41.76 -> 1.16 | Improved |
| replace FT with FT-SAM under Blended attack with 5% benign data | CIFAR-10 PreAct-ResNet18 · CIFAR-10 Blended 10% poison | ASR (%): 82.22 -> 4.91 | Improved |
| reduce benign data from 10% to 1% for FT-SAM (ablation) | FT-SAM · CIFAR-10 PreAct-ResNet18 | DER | No clear effect |
| replace FT with FT-SAM in the FP pruning pipeline (ablation) | Fine-pruning · CIFAR-10 CLA attack | ASR (%): 78.72 -> 33.76 | Improved |
| replace FT with FT-SAM after ANP masking (ablation) | ANP · CIFAR-10 Blended attack | ASR (%): 42.28 -> 2.57 | Improved |
| vary perturbation bound rho (smaller values) (ablation) | FT-SAM · CIFAR-10 complex attacks | ASR / DER | No clear effect |
Closest papers by meaning
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.92)
- Fusing Pruned and Backdoored Models: Optimal Transport-based Data-free Backdoor Mitigation (AAAI 2025 · similarity 0.88)
- Uncovering, Explaining, and Mitigating the Superficial Safety of Backdoor Defense (NeurIPS 2024 · similarity 0.88)
- From Toxic to Trustworthy: Using Self-Distillation and Semi-supervised Methods to Refine Neural Networks (AAAI 2024 · similarity 0.87)
- Backdoor Mitigation via Invertible Pruning Masks (NeurIPS 2025 · similarity 0.87)
- BAN: Detecting Backdoors Activated by Adversarial Neuron Noise (NeurIPS 2024 · 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.
