DHBE: Data-free Holistic Backdoor Erasing in Deep Neural Networks via Restricted Adversarial Distillation

Abstract

Backdoor attacks have emerged as an urgent threat to Deep Neural Networks (DNNs), where victim DNNs are furtively implanted with malicious neurons that could be triggered by the adversary. To defend against backdoor attacks, many works establish a staged pipeline to remove backdoors from victim DNNs: inspecting, locating, and erasing. However, in a scenario where a few clean data can be accessible, such pipeline is fragile and cannot erase backdoors completely without sacrificing model accuracy. To address this issue, in this paper, we propose a novel data-free holistic backdoor erasing (DHBE) framework. Instead of the staged pipeline, the DHBE treats the backdoor erasing task as a unified adversarial procedure, which seeks equilibrium between two different competing processes: distillation and backdoor regularization. In distillation, the backdoored DNN is distilled into a proxy model, transferring its knowledge about clean data, yet backdoors are simultaneously transferred. In backdoor regularization, the proxy model is holistically regularized to prevent from infecting any possible backdoor transferred from distillation. These two processes jointly proceed with data-free adversarial optimization until a clean, high-accuracy proxy model is obtained. With the novel adversarial design, our framework demonstrates its superiority in three aspects: 1) minimal detriment to model accuracy, 2) high tolerance for hyperparameters, and 3) no demand for clean data. Extensive experiments on various backdoor attacks and datasets are performed to verify the effectiveness of the proposed framework. Code is available at https://github.com/yanzhicong/DHBE

A Comprehensive Survey on Backdoor Attacks and Their Defenses in Face Recognition Systems

“…Access to synthesized data. Finally, 2 defense papers (see Table 4 and Table 5) applied to FRS models consider a defender who cannot assume any access to a clean dataset with respect to the task performed by a suspicious DNN [129], [130].…”

“…data poisoning). Instead, these defenses eschew data gathering and instead synthesize datapoints directly from the scrutinized DNN via model inversion [129] or distillation [130].…”

“…Doing so using clean data for instance may let a DNN defender yield a benign, equally-performing DNN, starting from a suspicious DNN. Such a method is found in the Data-free Holistic Backdoor Erasing (DHBE) defense [130], which generates a repaired model starting from a backdoored one. One advantage of DHBE is that it does not require the defender to have access to the internals of the suspicious DNN.…”