Regularized Adversarial Training (RAT) for Robust Cellular Electron Cryo Tomograms Classification

摘要: Cellular Electron Cryo Tomography (CECT) 3D imaging has permitted biomedical community to study macromolecule structures inside single cells with deep learning approaches. Many learning-based methods have since been developed classify from tomograms high accuracy. However, several recent studies demonstrated the lack of robustness in these models against often-imperceptible, designed changes input. Therefore, making existing subtomogram-classification robust remains a serious challenge. In this paper, we state-of-the-art subtomogram classifier on CECT images and propose method called Regularized Adversarial Training (RAT) defend wide range threats. Our results show that RAT improves for image classification over previous methods.