News for hackers. Stuff that matters.

A comment on Cellular Automata as Convolutional Neural Networks. Parent comment

Re:Cellular Automata as Convolutional Neural Networks by juskrey on Thursday August 13, 2020 @03:08AM (#24140733)
Indeed, CA can be represented by simple combinations of boolean functions, obviously by NN also, which is a combination of similar nonlinear functions.
[ Parent | on HN ]
Re:Cellular Automata as Convolutional Neural Networks by benibraz on Thursday August 13, 2020 @03:55AM (#24140978)
A wide enough NN can represent any arbitrary binary function, but it's not obvious that one can learn it.
[ Parent | on HN ]
Re:Cellular Automata as Convolutional Neural Networks by labelbias on Thursday August 13, 2020 @04:23AM (#24141119)
Yet the best NNs are deep, not wide.
[ Parent | on HN ]
Re:Cellular Automata as Convolutional Neural Networks by benibraz on Thursday August 13, 2020 @05:01AM (#24141311)
What do you mean by 'the best'? Deeper architectures are popular because they quiet easy to train. They do work well in practice on many tasks (especially vision) but they have their limits.

Infinite wide networks are a newly active field and has recently shown some promising results, theoretically [1, 2] and empirically [3].

[1] https://arxiv.org/abs/2001.06931 [3] https://arxiv.org/abs/1806.07572 [2] https://ai.googleblog.com/2020/03/fast-and-easy-infinitely-w...

[ Parent | on HN ]