Built-in Elastic Transformations for Improved Robustness

Abstract

We focus on building robustness in the convolutions of neural visual classifiers, especially against natural perturbations like elastic deformations, occlusions and Gaussian noise. Existing CNNs show outstanding performance on clean images, but fail to tackle naturally occurring perturbations. In this paper, we start from elastic perturbations, which approximate (local) view-point changes of the object. We present elastically-augmented convolutions (EAConv) by parameterizing filters as a combination of fixed elastically-perturbed bases functions and trainable weights for the purpose of integrating unseen viewpoints in the CNN. We show on CIFAR-10 and STL-10 datasets that the general robustness of our method on unseen occlusion, zoom, rotation, image cut and Gaussian perturbations improves, while significantly improving the performance on clean images without any data augmentation.

References in corpus (11)

• Wide Residual Networks
• Towards Deep Learning Models Resistant to Adversarial Attacks
• Provable defenses against adversarial examples via the convex outer adversarial polytope
• Robustness May Be at Odds with Accuracy
• Exploring the Landscape of Spatial Robustness
• Why do deep convolutional networks generalize so poorly to small image transformations?
• Interpreting Adversarially Trained Convolutional Neural Networks
• Spatial Transformer Networks
• Scale-Equivariant Steerable Networks
• Learning perturbation sets for robust machine learning
• Natural Perturbed Training for General Robustness of Neural Network Classifiers