News for hackers. Stuff that matters.
Re:Deep Learning: A Critical Appraisal (2018) by yters on Saturday May 04, 2019 @11:16AM (#19826944)
The question is whether those larger NNs are reaching an asymptotic limit that is way below known human capabilities. Currently, NNs can only be said to surpass human ability within a very narrow definition. However, if we do an apples to apples comparison, humans are still vastly superior to any NN on any task in terms of computation operations, memory and energy usage. We also learn in a inherently different way, using principles that are widely applicable. NNs, on the other hand, "learn" more in the sense of memorizing answers to a test. There is no understanding that NNs achieve which can be generalized and applied widely.
Re:Deep Learning: A Critical Appraisal (2018) by bArray on Sunday May 05, 2019 @06:42AM (#19832143)
It is of my opinion that one approach to tackle this could be to impose structure, such as "this NN is responsible for this part". That way we could speed up overall training and build more complex NN architectures.

That said I think there is still a lot of research to be done, they are not perfect, but I think we are barking up the right tree.

Re:Deep Learning: A Critical Appraisal (2018) by yters on Friday May 10, 2019 @01:54PM (#19880410)
Why do you think we are barking up the right tree? The assumption that mind = NN seems wrong on many levels.