Ryan Pyle

Local Convolutions Cause an Implicit Bias towards High Frequency Adversarial Examples

2020 · arXiv (Cornell University)

Adversarial Attacks are still a significant challenge for neural networks. Recent work has shown that adversarial perturbations typically contain high-frequency features, but the root cause of this phenomenon remains unknown. Inspired by theoretical work on …

A Quantitative Approach to Predicting Representational Learning and Performance in Neural Networks

2023 · arXiv (Cornell University)

A key property of neural networks (both biological and artificial) is how they learn to represent and manipulate input information in order to solve a task. Different types of representations may be suited to different …