I would add "in a way that you often can't predict or even detect or sometime even detect yourself"
"To learn from data, you have to make assumptions about it. These assumptions define what can be learned. As such, the structure of your hypothesis space—the architecture of your model—is extremely important. It encodes [...] the prior knowledge that the model starts with" (DLwP)
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Isn't this similar to assigning random weights in a neural network ??
-
Similar to picking the architecture of your net. With an ideal optimized those initial weights wouldn't matter. Though in practice they seem to (see Lottery Ticket Hypothesis).
- Show replies
New conversation -
-
-
And if all you assume is smoothness, you can only generalize locally.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Mais la meilleure façon de créer un model de qualité dépend fixement de comment se définit le dataset à entrainer non?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Can we estimate a model structure from the data? There was some work done before, but in reality we have to put our assumptions before training into the model.
-
Data itself is nothing without hypothesis
End of conversation
New conversation -
-
-
Isn't this what the No Free Lunch thorem also says !!
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.