And still more trouble for deep learning: “human visual system .... more robust [than deep neural nets] to nearly all of [a dozen] tested image manipulations” https://arxiv.org/abs/1808.08750
-
Well
@GaryMarcus, why do you think this is a problem with "deep learning" as opposed to a shortcoming in our current deep learning networks? I predict these problems will be completely solved within the DL paradigm.
-
-
It's bottom-up all the way down...
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
On the contrary, DL starts from the top. Backpropogation is very much top down as its name suggests.
- 1 more reply
New conversation -
-
-
Or may be because we have, so far, modeled only one area of the brain in DNN?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This Tweet is unavailable.
-
But the visual structure (Cornea to cortex) evolved bottom up. Contextual visual reasoning is emergent, not hierarchically imposed
End of conversation
-
-
-
Generalizing across data distributions is the next logical challenge for DL after (status quo) generalizing across samples from the same data distribution.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
It's funny watching Marcus pontificate about the flaws of deep learning (smart people knew DL was a red herring since the 90s) and cluelessly calling for GOFAI priors while completely ignoring the two most important components of intelligence: timing and yin-yang complementarity.
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.