MIRST-DM: Multi-instance RST with Drop-Max Layer for Robust Classification of Breast Cancer

Robust self-training (RST) can augment the adversarial robustness of image classification models without significantly sacrificing models' generalizability. However, RST and other state-of-the-art defense approaches failed to preserve generalizability reproduce their good on small medical sets. In this work, we propose Multi-instance with a drop-max layer, namely MIRST-DM, which involves sequence iteratively generated instances during training learn smoother decision boundaries datasets. The proposed layer eliminates unstable features helps representations that are robust perturbations. approach was validated using breast ultrasound dataset 1,190 images. results demonstrate achieves against three prevalent attacks.