It's not electric sheep. Scientists publish images of robots' dreams.

[DetlefK]

http://gizmodo.com/these-are-the-incredible-day-dreams-of-artificial-neura-1712226908
http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html

A neuron is a very simple tool: It receives a signal, magnifies or weakens it, and then passes it on. But working together in a neural network, they can perform complex mental operations.

Your brain is a neural network. So, scientists asked: "What is day-dreaming like for a computer based on a neural network architecture?"

Google trains artificial neural networks to recognize objects in images they see. And to simplify things, they only train these artificial "brains" to recognize one specific kind of objects. Here's what you get when you ask such a "brain" to give you a generic expression of what they are working with:



And when you have a neural network that is trained to recognize objects just look at the sky...



Scientists call this "inceptionism".

Now, what happens when you take the images the neural network has dreamed up and present it to him as images of real-life objects?

You get a feedback-loop.

And images like this:









--------------------------------------

Those are just still images. Imagine a neural network that is trained to recognize motions. Its "dreams" would be animations...

[enough]

1. Fascinating, thanks. (nt)

[tridim]

2. Art!

I believe most of life is our interpretation of the feedback loops in our heads.

[Fairgo]

3. Amazing!

Thank you!