Systematic Evaluation of Backdoor Data Poisoning Attacks on Image Classifiers
This work addresses safety risks for machine learning systems by assessing vulnerabilities to backdoor attacks, but it is incremental as it builds on prior research with a broader experimental scope.
The paper systematically evaluated backdoor data poisoning attacks on image classifiers, finding that attack success rates vary widely based on factors like model architecture and trigger patterns, and that backdoors can be removed with minimal additional training on clean data without harming performance.
Backdoor data poisoning attacks have recently been demonstrated in computer vision research as a potential safety risk for machine learning (ML) systems. Traditional data poisoning attacks manipulate training data to induce unreliability of an ML model, whereas backdoor data poisoning attacks maintain system performance unless the ML model is presented with an input containing an embedded "trigger" that provides a predetermined response advantageous to the adversary. Our work builds upon prior backdoor data-poisoning research for ML image classifiers and systematically assesses different experimental conditions including types of trigger patterns, persistence of trigger patterns during retraining, poisoning strategies, architectures (ResNet-50, NasNet, NasNet-Mobile), datasets (Flowers, CIFAR-10), and potential defensive regularization techniques (Contrastive Loss, Logit Squeezing, Manifold Mixup, Soft-Nearest-Neighbors Loss). Experiments yield four key findings. First, the success rate of backdoor poisoning attacks varies widely, depending on several factors, including model architecture, trigger pattern and regularization technique. Second, we find that poisoned models are hard to detect through performance inspection alone. Third, regularization typically reduces backdoor success rate, although it can have no effect or even slightly increase it, depending on the form of regularization. Finally, backdoors inserted through data poisoning can be rendered ineffective after just a few epochs of additional training on a small set of clean data without affecting the model's performance.