Disentangled Deep Autoencoding Regularization for Robust Image Classification
This addresses the robustness problem in deep learning for image classification, offering a complementary defense method against adversarial attacks, though it appears incremental as it builds on existing disentangling concepts.
The paper tackles the vulnerability of deep convolutional neural networks to adversarial attacks and poor generalization to novel test images by proposing a disentangled deep autoencoding regularization framework that learns separate appearance and geometric codes. The framework significantly outperforms traditional unregularized networks in robustness against adversarial attacks and generalization to novel test data across several benchmark datasets.
In spite of achieving revolutionary successes in machine learning, deep convolutional neural networks have been recently found to be vulnerable to adversarial attacks and difficult to generalize to novel test images with reasonably large geometric transformations. Inspired by a recent neuroscience discovery revealing that primate brain employs disentangled shape and appearance representations for object recognition, we propose a general disentangled deep autoencoding regularization framework that can be easily applied to any deep embedding based classification model for improving the robustness of deep neural networks. Our framework effectively learns disentangled appearance code and geometric code for robust image classification, which is the first disentangling based method defending against adversarial attacks and complementary to standard defense methods. Extensive experiments on several benchmark datasets show that, our proposed regularization framework leveraging disentangled embedding significantly outperforms traditional unregularized convolutional neural networks for image classification on robustness against adversarial attacks and generalization to novel test data.