The Machine is Dreaming

http://googleresearch.blogspot.com/2015/06/inceptionism-going-deeper-into-neural.html
https://photos.google.com/share/AF1QipPX0SCl7OzWilt9LnuQliattX4OUCj_8EP65_cTVnBmS1jnYgsGQAieQUc1VQWdgQ?key=aVBxWjhwSzg2RjJWLWRuVFBBZEN1d205bUdEMnhB
Google gives a neural network acid and asks it to return images. It combines dogs and fish, plants and buildings, birds and insects into a single images to optimize it's memory usage.

Fucking trippy.

This has already been done though:
https://www.youtube.com/watch?v=M2IebCN9Ht4

Say you want to know what sort of image would result in “Banana.” Start with an image full of random noise, then gradually tweak the image towards what the neural net considers a banana (see related work in [1], [2], [3], [4]). By itself, that doesn’t work very well, but it does if we impose a prior constraint that the image should have similar statistics to natural images, such as neighboring pixels needing to be correlated.

Their definition of ``work very well'' seems to be ``produce images that don't make neural networks look retarded''. I'm guessing 99% of the innovation here was crafting a prior constraint that worked to that effect.

Verdict: Just more smoke and mirrors from the neural net crowd.