DeepDream

DeepDream is a computer vision program that uses a convolutional neural network to find and amplify patterns in an image, producing dream-like, hallucinatory visual effects through a process its creators likened to algorithmic pareidolia. It was developed by Google engineer Alexander Mordvintsev, building on a deep convolutional network called Inception that had been built for the 2014 ImageNet Large-Scale Visual Recognition Challenge, and it was released publicly in July 2015. Rather than adjusting a network's weights, as in ordinary training, DeepDream keeps the trained weights fixed and instead repeatedly adjusts the pixels of an input image to increase the activation of chosen output neurons, such as those tuned to recognize faces or particular animals; after many iterations this produces the surreal, pattern-saturated imagery the program is known for. Google later released the underlying code as open source, leading to its wide adoption in web services, mobile apps and desktop software. This description is adapted from Wikipedia contributors under CC BY-SA 4.0; changes were made. https://creativecommons.org/licenses/by-sa/4.0/