Defense against Backdoor Attack on Pre-trained Language Models via Head Pruning and Attention Normalization

Xingyi Zhao, Depeng Xu, Shuhan Yuan
2/12/2026

No official code found. Best implementation: xingyizhao/PURE (11 stars, last commit 2026-02-22).

Semantic Scholar

Code Implementations(1)

xingyizhao/PURECommunity66%

Code associated with ICML (2024). "Defense against Backdoor Attack on Pre-trained Language Models via Head Pruning and Attention Normalization"

110PythonMay 26, 20248 months ago

Similar Papers

Connecting Pre-trained Language Model and Downstream Task via Properties of Representation

89% match2026

Not All Prompts Are Secure: A Switchable Backdoor Attack Against Pre-trained Vision Transfomers

81% match1 repo162026

Boosting 3D Single Object Tracking with 2D Matching Distillation and 3D Pre-training

79% match2026

POCA: Post-training Quantization with Temporal Alignment for Codec Avatars

79% match2026

Double-Filter: Efficient Fine-tuning of Pre-trained Vision-Language Models via Patch&Layer Filtering

79% match2026

Train Once and Explain Everywhere: Pre-training Interpretable Graph Neural Networks

79% match2026

Cite this paper

@article{zhao2026defense,
  title  = {Defense against Backdoor Attack on Pre-trained Language Models via Head Pruning and Attention Normalization},
  author = {Xingyi Zhao and Depeng Xu and Shuhan Yuan},
  year   = {2026},
  url    = {https://api.semanticscholar.org/CorpusID:5fbf503e2bd3b6930990ecd3851cf2c224409dff},
  journal = {ICML 2024 2024}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/5fbf503e2bd3b6930990ecd3851cf2c224409dff.svg)](https://codeofpaper.com/paper/5fbf503e2bd3b6930990ecd3851cf2c224409dff)

Discussion