"We put forward strong evidence that current deep learning methods are not yet sufficiently sample efficient when it comes to learning a language with compositional properties." Key word is yet, Gary.
Major news re #deeplearning & its limits: Yoshua Bengio’s lab confirms a key conclusion of Marcus 2001 and 2018: deep learning isn’t efficient enough with data to cope with the compositional nature of language. https://export.arxiv.org/abs/1810.08272
-
-
-
i have literally being hearing this reply for 20 years, since Marcus 1998. at what point do we acknowledge that might need other techniques, too, rather than deep learning alone?
- 3 more replies
New conversation -
-
-
I didn't read "current deep learning methods are not yet sufficiently sample efficient" as "confirms". Again this feels like criticising bricks for not being a city.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Very true. The capacity of current blends of
#AI stops at processing and executing simple commands and translating simplistic excerpts of text. I explored the topic it this week in@bdtechtalks https://bdtechtalks.com/2018/10/22/ai-deep-learning-human-language/ …#ArtificialIntelligence#DeepLearning -
agree with your article
@bendee983 state-of-the-art doesn't understand human language so is narryow. And it can't with the ML/DL methodologies deployed in Google Duplex. Our company does NLU which will create the step change needed.
End of conversation
New conversation -
-
-
Neat work by Bengio! Based on AlphaZero’s learning of ladder patterns (whole-board recurrent capture sequences) in Go though, learning compositionality is costly, but does happen with time. I wouldn’t exclude we can find shortcuts to get there faster, soon.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
funny thing is that "classical ai" and machine learning and linguistics never really went away in real-world problem solving and practical application; it's just been shadowed by recent progress and applicability of a +50 year research project. pendulum will shift as things do
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Any language (text, spoken, sign or body) is merely a representation of our thoughts that we want to convey. Till a system is not able to represent out thoughts and have capability to think (the true intelligence part), it wud not be able to solve anything close to human levels.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@lucy3_li this looks like your kind of stuffThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
language has no inherent meaning if not coupled with grounded experience. An upgraded NN won't "solve" the puzzle without understanding it.
-
That's why Bengio is trying to add something more to the word which is nothing but a marker for an experience.
- 5 more replies
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.