Boosting Adversarial Training with Hypersphere Embedding
This work addresses the problem of enhancing adversarial robustness for deep learning models, particularly in computer vision, by proposing an incremental improvement that integrates a lightweight module into existing adversarial training methods.
The paper tackles improving adversarial training for deep learning models by incorporating hypersphere embedding to regularize features onto compact manifolds, resulting in consistent robustness enhancements across multiple frameworks like PGD-AT and TRADES on datasets such as CIFAR-10 and ImageNet with minimal extra computation.
Adversarial training (AT) is one of the most effective defenses against adversarial attacks for deep learning models. In this work, we advocate incorporating the hypersphere embedding (HE) mechanism into the AT procedure by regularizing the features onto compact manifolds, which constitutes a lightweight yet effective module to blend in the strength of representation learning. Our extensive analyses reveal that AT and HE are well coupled to benefit the robustness of the adversarially trained models from several aspects. We validate the effectiveness and adaptability of HE by embedding it into the popular AT frameworks including PGD-AT, ALP, and TRADES, as well as the FreeAT and FastAT strategies. In the experiments, we evaluate our methods under a wide range of adversarial attacks on the CIFAR-10 and ImageNet datasets, which verifies that integrating HE can consistently enhance the model robustness for each AT framework with little extra computation.