Backdoor Policy

Backdoor attacks on machine learning models involve subtly poisoning training data to trigger malicious behavior during inference, compromising model integrity without obvious signs of tampering. Current research focuses on developing robust defense mechanisms, exploring techniques like unlearning poisoned samples, optimizing model parameters to smoother minima (e.g., using Fisher Information), and leveraging adversarial training to mitigate backdoor effects across various model architectures, including CNNs and transformers, and diverse data modalities (image, audio, text, video). Understanding the relationship between task similarity and backdoor detectability is a key area of investigation, with implications for improving both attack and defense strategies, ultimately enhancing the security and trustworthiness of machine learning systems.

Papers

September 29, 2024

Efficient Backdoor Defense in Multimodal Contrastive Learning: A Token-Level Unlearning Method for Mitigating Threats
Kuanrong Liu, Siyuan Liang, Jiawei Liang, Pengwen Dai, Xiaochun Cao
Contrastive Learning Backdoor Learning Backdoor Effect Input Level Backdoor Detection Backdoor Policy

September 1, 2024

Fisher Information guided Purification against Backdoor Attacks
Nazmul Karim, Abdullah Al Arafat, Adnan Siraj Rakin, Zhishan Guo, Nazanin Rahnavard
Backdoor Attack Backdoor Defense Fisher Information Backdoor Removal Backdoor Policy Backdoor Purification

March 29, 2024

A Backdoor Approach with Inverted Labels Using Dirty Label-Flipping Attacks
Orson Mengara
Backdoor Attack New Attack Backdoor Policy Label Flipping

November 13, 2023

Mitigating Backdoors within Deep Neural Networks in Data-limited Configuration
Soroush Hashemifar, Saeed Parsa, Morteza Zakeri-Nasrabadi
Deep Neural Network DNN Framework Limited Data Mitigating Backdoor Backdoor Policy

July 20, 2023

Shared Adversarial Unlearning: Backdoor Mitigation by Unlearning Shared Adversarial Examples
Shaokui Wei, Mingda Zhang, Hongyuan Zha, Baoyuan Wu
Adversarial Example Backdoor Attack Adversarial Risk Backdoor Policy

June 30, 2023

Efficient Backdoor Removal Through Natural Gradient Fine-tuning
Nazmul Karim, Abdullah Al Arafat, Umar Khalid, Zhishan Guo, Naznin Rahnavard
Backdoor Defense Deep Neural Network Parameter Backdoor Removal Loss Surface Backdoor Policy Backdoor Purification

March 25, 2023

Backdoor Attacks with Input-unique Triggers in NLP
Xukun Zhou, Jiwei Li, Tianwei Zhang, Lingjuan Lyu, Muqiao Yang, Jun He
Natural Language Processing Backdoor Attack Backdoor Trigger Backdoor Injection Trigger Design Backdoor Policy

November 18, 2022

Provable Defense against Backdoor Policies in Reinforcement Learning
Shubham Kumar Bharti, Xuezhou Zhang, Adish Singla, Xiaojin Zhu
Reinforcement Learning Active Subspace Provable Defense Backdoor Policy

October 12, 2022

Understanding Impacts of Task Similarity on Backdoor Attack and Detection
Di Tang, Rui Zhu, XiaoFeng Wang, Haixu Tang, Yi Chen
Data Detection Backdoor Attack Impact Assessment Backdoor Detection Task Similarity Stealthy Backdoor Attack Backdoor Policy