About SARM. At this point I am 100% convinced that the VGG16 experiment is not for real. Most likely a big experimental mistake, not fraud.
-
That said. If this paper manages to reignite interest in sparse coding, then that's a positive result. Important, under-appreciated topic.
Backprop is computationally inefficient and data-inefficient. In the future we won't be using it. But its replacement has yet to be found.
-
-
@withfries2 aren't people working with synthetic gradients nowadays? -
maybe that's what
@ML_Hipster is up to these days
End of conversation
New conversation -
-
-
It'll be interesting to see how this all pans out. I think it's real though
-
it's difficult to parse what the paper is talking about, but if my interpretation is correct then I know what I'm talking about.
- Show replies
New conversation -
-
-
@withfries2 The problem isn't with backprop, it's with the training paradigm. E.g., end-to-end from image to human category label.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
"this is the best bad idea we have sir, by far" https://www.youtube.com/watch?v=2zs-41GCI9E …
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
don't see problem in backprop, i.e., derivatives. The problem is probably on using 1st order algs. Maybe going to 2nd order helps
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
precisely for the refinement of the parameters, like any optimization. Backprop is good for the topological flexibility of DNNs
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@luciotorre target propagation is an interesting optionThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
pretty much like democracy.
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.