"representationally efficient" meaning here that it introduces modularity and hierarchy in the representations.
Convnets are a representationally-efficient way to do pattern recognition on data with continuous axes (time, space). No magic whatsoever. https://twitter.com/davidsuculum/status/854824527896993792 …
This Tweet is unavailable.
-
-
-
Which is how you make *any* system more efficient.
- Show replies
New conversation -
-
-
A NN architecture is basically just a set of assumption about properties of your data. CNNs assume translation invariance and modularity.
-
Whether an architecture works well on a problem depends on the match between actual properties and assumptions. There is no universality.
End of conversation
New conversation -
-
-
Comme je t'aime, Francois Chollet!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
do convnets get around "free lunch theorem"? "discovering features" is awesome, but convnets also have to deal with very high dims
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.