Learning from almost nothing: How neural networks survive heavy input corruption

In three linesNeural networks remain robust against >90% input corruption, far beyond human recognition. Mean-field analysis of infinite-width networks reveals MLPs implement a universal prototype rule (nearest-class-mean) across architectures, explaining why learning succeeds despite massive noise.

Summary generated by Claude — human-verified