Itay Safran

A Simple Explanation for the Existence of Adversarial Examples with Small Hamming Distance

2019 · arXiv (Cornell University)

The existence of adversarial examples in which an imperceptible change in the input can fool well trained neural networks was experimentally discovered by Szegedy et al in 2013, who called them "Intriguing properties of neural …

Random Shuffling Beats SGD Only After Many Epochs on Ill-Conditioned\n Problems

2021 · arXiv (Cornell University)

Recently, there has been much interest in studying the convergence rates of\nwithout-replacement SGD, and proving that it is faster than with-replacement\nSGD in the worst case. However, known lower bounds ignore the problem's\ngeometry, including its condition …