Neural Variability Enhances Artificial Network Robustness

In three linesResearchers demonstrate that structured noise in artificial neural network activations improves robustness against adversarial attacks and naturalistic image modifications. This approach is inspired by neural variability observed in biological cortex and requires only local information.

Summary generated by Claude — human-verified