Hacker News new | ask | show | jobs
by the8472 3598 days ago
That's what deep dream was about to some extent.

They tweak the input to maximize the response of specific neurons somewhere in the middle of the network to figure out what those neurons actually "learned".