Countering Backdoor Attacks in Image Recognition: A Survey and Evaluation of Mitigation Strategies

The widespread adoption of deep learning across various industries has introduced substantial challenges, particularly in terms of model explainability and security. The inherent complexity of deep learning models, while contributing to their effectiveness, also renders them susceptible to adversarial attacks. Among these, backdoor attacks are especially concerning, as they involve surreptitiously embedding specific triggers within training data, causing the model to exhibit aberrant behavior when presented with input containing the triggers. Such attacks often exploit vulnerabilities in outsourced processes, compromising model integrity without affecting performance on clean (trigger-free) input data. In this paper, we present a comprehensive review of existing mitigation strategies designed to counter backdoor attacks in image recognition. We provide an in-depth analysis of the theoretical foundations, practical efficacy, and limitations of these approaches. In addition, we conduct an extensive benchmarking of sixteen state-of-the-art approaches against eight distinct backdoor attacks, utilizing three datasets, four model architectures, and three poisoning ratios. Our results, derived from 122,236 individual experiments, indicate that while many approaches provide some level of protection, their performance can vary considerably. Furthermore, when compared to two seminal approaches, most newer approaches do not demonstrate substantial improvements in overall performance or consistency across diverse settings. Drawing from these findings, we propose potential directions for developing more effective and generalizable defensive mechanisms in the future.

Unlearning Backdoor Attacks through Gradient-Based Model Pruning

In the era of increasing concerns over cybersecurity threats, defending against backdoor attacks is paramount in ensuring the integrity and reliability of machine learning models. However, many existing approaches require substantial amounts of data for effective mitigation, posing significant challenges in practical deployment. To address this, we propose a novel approach to counter backdoor attacks by treating their mitigation as an unlearning task. We tackle this challenge through a targeted model pruning strategy, leveraging unlearning loss gradients to identify and eliminate backdoor elements within the model. Built on solid theoretical insights, our approach offers simplicity and effectiveness, rendering it well-suited for scenarios with limited data availability. Our methodology includes formulating a suitable unlearning loss and devising a model-pruning technique tailored for convolutional neural networks. Comprehensive evaluations demonstrate the efficacy of our proposed approach compared to state-of-the-art approaches, particularly in realistic data settings.