@ylecun i absolutely do not dismiss gradient based learning (as part of larger system) & have said so publicly and to you privately. please read deep learning: a critical appraisal, particularly parts about deep learning as tool of larger system, and stop misrepresenting me. 1/2https://twitter.com/ylecun/status/1065837329221328896 …
-
Show this thread
-
Replying to @GaryMarcus
As I, Geoff, Léon Bottou and others have said: "replace symbols by vectors and logic by algebra". This does not preclude reasoning, or even variable binding. But it may make reasoning differentiable. Read Leon's "from ML to Machine Reasoning". Call it symbolic or not, no matter.
6 replies 8 retweets 42 likes -
Replying to @ylecun @GaryMarcus
Serious question: a) must reasoning be differentiable to be called reasoning? and b) that notwithstanding, do you assert that there are not other paths to what we would call reasoning? (And I’m searching for papers that help me understand your POV)
2 replies 1 retweet 7 likes -
We generally convert both learning and reasoning into optimization. We can optimize with gradient-based methods if the representations are differentiable. There are other methods, but stochastic gradient descent has proved to be extremely successful
3 replies 4 retweets 16 likes
why would that make me “half right” when it is what i am calling for?
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.