Shot Backdoor

Shot backdoor attacks exploit vulnerabilities in machine learning models, particularly those trained with limited data (few-shot learning), to manipulate their predictions by embedding hidden triggers. Current research focuses on developing both more effective attacks, leveraging techniques like neural tangent kernels and adversarial prompt tuning, and robust defenses, such as Shapley value-based neuron pruning. Understanding and mitigating these attacks is crucial for ensuring the security and reliability of machine learning systems across diverse applications, including natural language processing, image recognition, and visual object tracking.

Papers

June 6, 2024

PromptFix: Few-shot Backdoor Removal via Adversarial Prompt Tuning
Tianrong Zhang, Zhaohan Xi, Ting Wang, Prasenjit Mitra, Jinghui Chen
Language Model Natural Language Processing Natural Language Processing Model Backdoor Defense Shot Backdoor

December 31, 2023

Does Few-shot Learning Suffer from Backdoor Attacks?
Xinwei Liu, Xiaojun Jia, Jindong Gu, Yuan Xun, Siyuan Liang, Xiaochun Cao
Backdoor Attack Shot Learning Shot Backdoor

October 12, 2022

Few-shot Backdoor Attacks via Neural Tangent Kernels
Jonathan Hayase, Sewoong Oh
Deep Neural Network Backdoor Attack Neural Tangent Kernel Poisoning Attack Attack Success Rate Shot Backdoor

July 25, 2022

Technical Report: Assisting Backdoor Federated Learning with Whole Population Knowledge Alignment
Tian Liu, Xueyang Hu, Tao Shu
Federated Learning Technical Report Backdoor Defense Knowledge Alignment Backdoor Effect Shot Backdoor

January 31, 2022

Few-Shot Backdoor Attacks on Visual Object Tracking
Yiming Li, Haoxiang Zhong, Xingjun Ma, Yong Jiang, Shu-Tao Xia
Backdoor Attack Visual Object Tracking Hidden Backdoor Speaker Tracking Shot Backdoor

December 30, 2021

Few-shot Backdoor Defense Using Shapley Estimation
Jiyang Guan, Zhuozhuo Tu, Ran He, Dacheng Tao
Deep Neural Network Backdoor Attack Shapley Value Pruning Method Shot Backdoor