GAT: Guided Adversarial Training with Pareto-optimal Auxiliary Tasks
This addresses the challenge of adversarial robustness for machine learning models, particularly in resource-constrained settings like medical imaging, though it is incremental as it builds on existing adversarial training methods.
The paper tackles the problem of improving adversarial robustness without incurring high data collection and computation costs by proposing Guided Adversarial Training (GAT), which uses auxiliary tasks to enhance training; it increases robust AUC on CheXpert from 50% to 83% and achieves 56.21% robust accuracy on CIFAR-10.
While leveraging additional training data is well established to improve adversarial robustness, it incurs the unavoidable cost of data collection and the heavy computation to train models. To mitigate the costs, we propose Guided Adversarial Training (GAT), a novel adversarial training technique that exploits auxiliary tasks under a limited set of training data. Our approach extends single-task models into multi-task models during the min-max optimization of adversarial training, and drives the loss optimization with a regularization of the gradient curvature across multiple tasks. GAT leverages two types of auxiliary tasks: self-supervised tasks, where the labels are generated automatically, and domain-knowledge tasks, where human experts provide additional labels. Experimentally, GAT increases the robust AUC of CheXpert medical imaging dataset from 50% to 83% and On CIFAR-10, GAT outperforms eight state-of-the-art adversarial training and achieves 56.21% robust accuracy with Resnet-50. Overall, we demonstrate that guided multi-task learning is an actionable and promising avenue to push further the boundaries of model robustness.