News for hackers. Stuff that matters.
Re:What intelligent machines need to incorporate from the ne... by return0 on Saturday June 03, 2017 @03:58AM (#14475469)
I think its fair to say that Hinton's justifications are post-hoc. I think a lot of neural network scientists also do that, as there are many basic components that dont have a biological analogue or even contradict biology, like backpropagation, LSTMs, RBM training layer-by-layer, semilinear activation functions. I can't think of an instance for example where the designers made a choice of connectivity "because that is how the brain does it" and they ended up having a better performing network. There have been recently attempts to match ANNs with spiking networks (e.g. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5021692/), but only rough sketches of ideas are provided.

There are no comments on this comment yet.