Do computers dream about fractal antilopes ?http://i.imgur.com/jZtbz7f.png
Inceptionism: Going Deeper into Neural Networks
81–90 of 166 posts
Re: Inceptionism: Going Deeper into Neural Networks
#82The reason they look so 'fractal-like' (e.g. trippy!) is because they actually are fractals! In the same way a normal fractal is a recursive application of some drawing function, this is a recursive application of different generation or "recognition -> generation" drawing functions built on top of the CNN. So I believe that, given a random noise image, these networks don't generate the crazy trippy fractal patterns…
> The reason they look so 'fractal-like' (e.g. trippy!) is because they actually are fractals! Do they exhibit self-similarity at different zoom levels?
I believe they do (in the sense that if you take one of these images, zoom it in, and run it through the algorithm again, it'll take the micro-features of the animals it hallucinated and hallucinate more animals on top of them).
Re: Inceptionism: Going Deeper into Neural Networks
#83I'll repeat what I posted on facebook because I thought it was clever: "Yes, but only if we tell them to dream about electric sheep." So, tell the machine to think about bananas, and it will conjure up a mental image of bananas. Tell it to imagine a fish-dog and it'll do its best. What happens if/when we have enough storage to supply it a 24/7 video feed (aka eyes), give a robot some navigational logic (or strap it t…
It would have some kind of intelligence, at least able to recall information and form associations between things. But there's no reason to think that it would come out looking human. I mean you can show a dog lots and lots of images and it doesn't turn human.
Re: Inceptionism: Going Deeper into Neural Networks
#84Really cool. You could generate all kinds of interesting art with this. I can't help but think of people who report seeing faces in their toast. Humans are biased towards seeing faces in randomness. A neural network trained on millions of puppy pictures will see dogs in clouds.
That's essentially precisely what's happening here. You can see in the different pictures where different sets of training data were used---buildings, faces, animals.
Give the machine millions of reference images to work from and then tell it to find those images in noise, and it will succeed (because it literally can't "imagine" anything else for the noise to be).
Re: Inceptionism: Going Deeper into Neural Networks
#85This story has been picked up by The Guardian: http://www.theguardian.com/technology/2015/jun/18/google-ima...
Re: Inceptionism: Going Deeper into Neural Networks
#86Am I the only person who is not entirely happy about the overuse of the pop-culture term 'inception' for everything that is remotely nested, recursive or strange-loop-like? In this paper, we will focus on an efficient deep neural network architecture for computer vision, codenamed Inception, which derives its name from the Network in network paper by Lin et al [12] in conjunction with the famous “we need to go deeper…
I haven't personally noticed any buzz-wordiness about that term lately. Maybe I'm just not looking at the same stuff as you.
In other words, you haven't trained your personal neural net with enough Inception-meme instances to be finding it everywhere in the noise? ;)
Re: Inceptionism: Going Deeper into Neural Networks
#87Earlier quoted context omitted.
There's also a few sequels: http://ansible.uk/writing/c-b-faq.html http://www.lightspeedmagazine.com/fiction/different-kinds-of... And What Happened at Cambridge IV , which I can't find online.
Here's What Happened at Cambridge IV: https://books.google.co.uk/books?id=5d9hHvD-T7gC&lpg=PA264&o... This last story makes the ML images even more disturbing! Highly recommend these stories. They'd make a great black mirror episode.
Google Books previews appears to block pages randomly per user --- I don't see the entire text, unfortunately.
Re: Inceptionism: Going Deeper into Neural Networks
#88The reason they look so 'fractal-like' (e.g. trippy!) is because they actually are fractals! In the same way a normal fractal is a recursive application of some drawing function, this is a recursive application of different generation or "recognition -> generation" drawing functions built on top of the CNN. So I believe that, given a random noise image, these networks don't generate the crazy trippy fractal patterns…
> The reason they look so 'fractal-like' (e.g. trippy!) is because they actually are fractals! Do they exhibit self-similarity at different zoom levels?
Yes, you can see zoomed-in sections of the last image in the article, about halfway down this gallery https://photos.google.com/share/AF1QipPX0SCl7OzWilt9LnuQliat...
Re: Inceptionism: Going Deeper into Neural Networks
#89Earlier quoted context omitted.
It reminds me of that parrot image that was said to crash human brains, only even more intense. I certainly experienced some effect, as while looking at it and trying to figure out what exactly it was, I felt my head heating up --- probably increased blood flow.
In case anyone hasn't read the story this is referring to: "BLIT" by David Langford. http://www.infinityplus.co.uk/stories/blit.htm
Wow. People thought the weirdest stuff about logic and brains back in the day.
Re: Inceptionism: Going Deeper into Neural Networks
#90a really early version of this:
http://draves.org/fuse/
published as open source in the early 90s.
not NN but does have the same image matching/searching.