Why does deep and cheap learning work so well?

Henry W. Lin, Max Tegmark, arXiv, Sept 10, 2016
Commentary by Stephen Downes

There's a good Technology Review summary of this article. In a nutshell: why do deep learning algorithms, which simulate neural networks, work so well? Mathematically, they should be much less effective, because they are attempting to select the best answer from an enormous number of possible outcomes. According to this paper, the reason is that the laws of physics are biased toward certain outcomes, and neural networks - which emulate physical processes - are biased in a similar manner. “We have shown that the success of deep and cheap learning depends not only on mathematics but also on physics, which favors certain classes of exceptionally simple probability distributions that deep learning is uniquely suited to model.” It's an important lesson: the universe may be described by mathematics, but it is not defined by mathematics.

Views: 0 today, 247 total (since January 1, 2017).[Direct Link]
Creative Commons License. gRSShopper

Copyright 2015 Stephen Downes ~ Contact: stephen@downes.ca
This page generated by gRSShopper.
Last Updated: Oct 23, 2017 09:25 a.m.