https://www.nature.com/articles/s41467-019-11786-6 … A very sensible article! If neural nets are to replicate animal intelligence, researchers will have to “decode” a relationship between network architectures and behaviors (the way CNNs are a network representation of translation invariance).
-
-
Of course, ML researchers *do* innovate! And I don’t have much confidence in my ability to forecast how fast they’ll solve the relevant remaining problems. But I am confident that the position “we just need more computing power and we’ll get human-level AI” is false.
Show this threadThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
When you say 'network architecture', what do you mean? More precisely, how broadly would you define 'network'? I ask as I'm not sure if this is meant as a strongly restricted subset of software or not, & if so then how exactly.
-
Network architecture = what weights are connected to what other weights, in an ANN. I think neural networks are a strict subset of software, yes.
- 5 more replies
New conversation -
-
-
@s_r_constantin well we know that with enough brutality a fully connected NN can do basically what a convnet does. I agree with your sentiment however I would caution that sometimes the machine does have the ability to brute force its way over the finish line. -
(as in both more data and more compute)
- 2 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.