Aleksandr Groznykh “in the cloud with data”
Deep Dream is backpropagation to the input image to maximize activations on a given layer of a neural network. In other words, amplifying whatever the network already sees.
When dreaming on deeper and deeper layers, we can explore how a neural network forms its inner representation of the world. We see complex features emerging from simple patterns.
We suggest to consider that our own minds work in a similar manner, constructing the illusion of conscious experience from a handful of sensed signals.
Made with lucid — a tensorflow library for neural network interpretability research
Colab notebook is available upon request (@algroznykh).
Special thanks to:
Dmitry Rotkin for making the awesome 3d space data center environment.
Lotus.eater for making the music from Saint-Petersburg data centers field recordings.
Text: provided by the author
Music: Drone Liberation Front