“Deep Learning: A Critical Appraisal”
Gary Marcus : https://arxiv.org/abs/1801.00631
#ArtificialIntelligence #DeepLearning #MachineLearningpic.twitter.com/YlsweQKsEl
-
For sure. No doubt it my mind that symbol manipulation is critical; and also that variously composing and reasoning about those symbols is a hard problem. I don't (yet!) see that, with appropriate structure and appropriate representations, it's impossible for NNs to do all that.
neural nets need to incorporate symbols, not deny them. it's not impossible, just mostly out of fashion.
-
-
I think you could argue that neural nets *do* incorporate symbols e.g. could not a Word2Vec vector be regarded as a symbol? The problem, IMO, is that these particular symbols are not easily composable (or decomposable) in meaningful ways, or generally possible to reason about.
-
2. Important question is whether deep learning alone supports such tracking and understanding. In Rebooting and Algebraic I have give many reasons to think otherwise.
- 1 more reply
New conversation -
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.