This article from medium.com's Backchannel portal is one of those descriptions of new creative effort you just have let simmer in your head for a while.
This is the one, of course, that chronicled Alexander Mordvintsev's inspired idea of letting convolutional neural nets go off on their own exploration of visual input, instead of simply trying to arrange the complex set of synaptic weighting, and eventual neuron firing, that would allow the whole to correctly identify desired image elements. To say that the results are stunning is to hardly do them justice at all.
Please do give this a read. It is fascinating, as well as stimulating a good number of delightful questions.
No comments:
Post a Comment