Understanding how adversarial noise affects single image classification

Abstract

In recent trends, computer vision applications have seen massive implementation of supervised learning with convolutional neural networks. In this paper, we have analyzed image classifiers and their classification accuracy. Also, we have measured their robustness upon introduction to various noise layers. Furthermore, we have implemented a generative adversarial network for the generator task of adversarial noise generation and the discriminator task of single image classification on the handwritten digits database. Our experiments are yielding progressive results and we have performed conditional and quantifiable evaluation of the generated samples.