Post #1727831
2026-04-08 20:34 UTC
@ly2en while we shouldn't think the layers are learning those specific facial features, I'd say it's a decent illustration for what they try to illustrate: the later the layer, the more semantic are the features that it optimises for.
Replies (2)
-
@ly2en@chaos.social 2026-04-10 05:36
@rtfb It's a good point, but somehow I still believe that this can be demonstrated without using oversimplified analogies.
-
@ly2en@chaos.social 2026-04-10 19:09
@rtfb also, can your statement be mathematically proven?