Explaining Away Attacks Against Neural Networks

Sean Saito, Jin Wang

We investigate the problem of identifying adversarial attacks on image-based neural networks. We present intriguing experimental results showing significant discrepancies between t…