What a Deep Neural Network thinks about your #selfie

Andrej Karpathy, Andrej Karpathy's Blog, Oct 28, 2015
Commentary by Stephen Downes
files/images/grid_render_posneg.jpg

This article is a reaally good indication of thee sort of thing that can be done with a neural network (I hesitate to classify these as deep neural networks, because as the article makes clear, they depend on training sets). It shows how unpredictable the science remains. The type of network described in the story, Convolutional Neural Networks, were first developed back in the 1980s (this is also around the same time Rumelhart and McClelland's Parallel Distributed Processing came out). ConvNets, as they were known, remained a footnote until 2012, when computer power finally caught up to the process. Then they were everywhere, doing things like recognizing numbers and telling people their selfies are ugly. The article is worth a read because it gives you a good sense of how the algorithms are used on the particular problem of selfie classification, which should offer a sense of their wider application.

Views: 0 today, 175 total (since January 1, 2017).[Direct Link]