Robustness Testing Framework For Neural Network Image Classifier

Abstract Neural network has made remarkable achievements in the field of image classification, but they are threatened by adversarial examples process application, making robustness neural classifiers face danger. Programs or software based on need to undergo rigorous testing before use and promotion, order effectively reduce losses security risks. To comprehensively test standardize process, starting from two aspects generated content interference intensity, a variety sets constructed, framework suitable for is proposed. And feasibility effectiveness method verified LENET-5 model reinforced adversavial training.