Thanks! As one of the people involved in DeepDream, I agree that was the most natural interpretation of our results -- but I think there were some alternatives. Since deepdream maximizes all activations, you could image a kind of "puzzle pieces" type situation... [CONT]
-
-
-
... where the network only detects local structure, but they natural way for the local structure to tile together emulates global structure. Later work on feature visualization ruled that out, but there are a bunch of subtle other possibilities.
- 1 more reply
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.