Hacker News new | ask | show | jobs
by fbdab103 1070 days ago
Famously, some of the image models were so overtrained they could still yield impressive results if the colors were removed.
1 comments

This wasn't so much overtraining, as the models learning something different than what we expected. If you look at a pixel by pixel representation of an image, textures tend to be more significant/unique patterns than shapes. There are some funny studies from the mid 2010s exploring this.