I made my own post about this before I saw this one. I'm just going to copy and paste the body here. What happens if you run an aritificial neural network in reverse? It makes art.
http://www.atlasobscura.com/articles/this-mystery-photo-haunting-reddit-appears-to-be-image-recognition-gone-very-weird Or maybe you could say that it dreams. In-depth discussion of the Google team here: https://news.ycombinator.com/item?id=9736598 High resolution images, suitable for desktop background here: https://photos.google.com/share/AF1QipPX0SCl7OzWilt9LnuQliattX4OUCj_8EP65_cTVnBmS1jnYgsGQAieQUc1VQWdgQ?key=aVBxWjhwSzg2RjJWLWRuVFBBZEN1d205bUdEMnhB
This begs the question, "do androids dream of electric sheep?" This article was amazing. I'd never thought of this way of checking on the training of the network. Artifacts like a dumbbell connected to an arm is an excellent example of a feature of a training set that you're not immediately aware of, but is problematic nonetheless.Or maybe you could say that it dreams.