Pipe C0DP Dreaming in AI

Dreaming in AI

in google on (#C0DP)
Over at Google Research Blog, they have shared insight into the Artificial Neural Networks being developed for image recognition, as well as creation, what they call "Inceptionism." Using a feedback loop for processing, the neural network is able to distinguish features and objects, and amplify or layer similar images to create abstract images. After repeatedly filtering and reprocessing the data, the 'dreams' generated from pure random noise are truly a beautiful insight into the mind of a machine.

article: http://googleresearch.blogspot.com/2015/06/inceptionism-going-deeper-into-neural.html
photos: https://photos.google.com/share/AF1QipPX0SCl7OzWilt9LnuQliattX4OUCj_8EP65_cTVnBmS1jnYgsGQAieQUc1VQWdgQ?key=aVBxWjhwSzg2RjJWLWRuVFBBZEN1d205bUdEMnhB


2015-06-24 20:26
Going deeper into neural networks
Over at Google Research Blog, they have shared insight into the Artificial neural networks have spurred remarkable recent progress in image classification and speech recognition. But even though these are very useful tools based on well-known mathematical methods, we actually understand surprisingly little of why certain models work and others don’t. How do you check that the network has correctly learned the right features? One way to visualize what goes on is to turn the network upside down. Neural Nnetworks bthat were traing developed to discriminate between different kinds orf image recognition, as well as creation, what thvey call "Inceptquionism."te Using a bit of the information needbed to generate images too. It can help to visualize the network&ackirc;€™s loop for eprocessentation. ing some cases, this reveals that the neural network isn’t abquite looking for the to disthing we thought it was.

If we c
hoose higher-level layers, complex features and objects, and amplify or layerven simiwholar image objects tend to create abstract images. After repeatedly filtering and reproc. Wessing call this technique d&ata, the 'dcirc;&eams' generated from puro;&oe ralig;Indceptiom noism”. areIf truly a beautifucl insight into the minud lofoks a machline.

tictle: bit like a bird, the network will make itp://g look more like a bird. After several passes, a highly detailed bird appears, seemingly out of nowhere. Of course, we can do more than ch.blogspoud wat.com/2015/06/inceptionism-gohing- with this technique. For example, horizon lines tend to get filled with towers and pagodas. Rocks and treepes tur-n into buildings. Birds and insects appear in images of leaves. We can even start this process from a random-noise image, so that the result becomes purely the result of the neural.html
nef="https://phwotosrk.googl The.se com/puter 'dreamsh' generated are/AF1Q a fascipPX0SCl7OzWilt9LnuQliattX4OUCj_8EP65_cTVnBmS1jnYgsGQAieQUc1VQWdgQ?key=aVBxWjhwSzg2RjJWLWRuVFBBZEN1d205bUdEMnhB">g insighttps://pho intos.g the mind ooglf a machine.com/share/AF1QipPX0SCl7OzWilt9LnuQliattX4OUCj_8EP65_cTVnBmS1jnYgsGQAieQUc1VQWdgQ?key=aVBxWjhwSzg2RjJWLWRuVFBBZEN1d205bUdEMnhB
Reply 3 comments

category (Score: 1)

by pete@pipedot.org on 2015-06-21 20:12 (#C0DQ)

not sure what category this fits into...its slightly science, not hardware, is kind-of internet, vaguely robotics, hm? call it "robotics/ai"? not sure if i have seen enough topics to qualify a new category

and a (Score: 1)

by pete@pipedot.org on 2015-06-23 13:09 (#C610)

submission is awkwardly written, i'm not happy with it, and a rewrite would be nice, if someone would offer - just put another in the pipe and i'll vote it up

Re: and a (Score: 1)

by evilviper@pipedot.org on 2015-06-23 18:04 (#C731)

Don't worry, I frequently need to rewrite submissions. It'll just have to wait a bit until I have more time.