Just read Gary Marcus' "Deep Learning Appraisal" paper. He outlined challenges for Deep Learning systems: 1. Data Hungry 2. Shallow and limited capacity for transfer 3. no way to deal with hierarchical structure 4. struggled with open ended inference
-
Show this thread
-
5. not sufficiently transparent 6. not been well integrated with prior knowledge 7. cannot distinguish causation from correlation 8. assumes a largely stable world 9. works well for approximation but its answers cannot be fully trusted 10. difficult to engineer with
1 reply 0 retweets 0 likesShow this thread -
Mostly agreed, and experienced in first hand in my projects, but I wonder if the recent so-called "ImageNet moment" of NLP (with Elmo, Bert etc.) has changed
@GaryMarcus' view on 3 and 4.2 replies 0 retweets 1 likeShow this thread
Replying to @fatih_bulut
not so much, at least so far
10:51 AM - 26 Aug 2019
0 replies
0 retweets
1 like
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.