flexiverse
|
|
« on: July 04, 2015, 10:34:59 AM » |
|
|
|
|
Logged
|
|
|
|
kram1032
|
|
« Reply #1 on: July 04, 2015, 11:15:16 AM » |
|
This video was already linked in the other thread. Also, any reason to link it twice? Though it's definitely really cool to look at.
|
|
|
Logged
|
|
|
|
cKleinhuis
|
|
« Reply #2 on: July 04, 2015, 12:47:08 PM » |
|
i think its worth to mention it in its own thread, the double postings seems to be a bug in the implementation of my forums embed video hack, which does embed linked tags as well will try to fix that regarding the art form, what we see here is more or less just what it says, the outputs of a chaotical machine, this machine is by nature chaotic, due to its many many many process of self application, the overall system that is build here and interconnected to produce such an output is quite astonishing to see what levels we have reached
|
|
|
Logged
|
---
divide and conquer - iterate and rule - chaos is No random!
|
|
|
flexiverse
|
|
« Reply #3 on: July 05, 2015, 02:54:37 AM » |
|
i think its worth to mention it in its own thread, the double postings seems to be a bug in the implementation of my forums embed video hack, which does embed linked tags as well will try to fix that regarding the art form, what we see here is more or less just what it says, the outputs of a chaotical machine, this machine is by nature chaotic, due to its many many many process of self application, the overall system that is build here and interconnected to produce such an output is quite astonishing to see what levels we have reached I think the key thing here, is that is really is a new art form. This has nothing to do with fractals. It's neural nets trained to recognise images, which is a completely and utterly light years different to how fractals are programmed. Some of the data sets have millions of images. So training your own net with specific images is going to be a tough call. So it looks like this deep mind inceptionism is going to create a whole new "thing". It's really fascinating when a new "thing" takes off. The only last big thing visually was 3d fractals like the mandel box ....
|
|
|
Logged
|
|
|
|
kram1032
|
|
« Reply #4 on: July 05, 2015, 03:15:38 AM » |
|
It's still pretty related to fractal processes, I'd say. The technique is very different but those images happen via feedback loops and the zoom sequence pretty much is like fractal deep zooms (except that this technique isn't determinnistic so each time you'll get a different zoom path). In the end it's still a process finding and rendering attractors. Those attractors just happen to have a LOT of information which makes them look like pieces of real life objects.
|
|
|
Logged
|
|
|
|
flexiverse
|
|
« Reply #5 on: July 06, 2015, 03:58:46 AM » |
|
It's still pretty related to fractal processes, I'd say. The technique is very different but those images happen via feedback loops and the zoom sequence pretty much is like fractal deep zooms (except that this technique isn't determinnistic so each time you'll get a different zoom path). In the end it's still a process finding and rendering attractors. Those attractors just happen to have a LOT of information which makes them look like pieces of real life objects.
For sure it's fractal related. But the process isn't really feedback loops/attractors. Infact I don't think anyone understands how internally these nets work. Once say you train a net to recognise dogs, from millions of dog images, it can recognise a dog. So if it sees something that looks like an eye, it tries to fit and eye there. So the process we are seeing is the artificial brain cells doing their job. Reading how how these nets work, well it's complicated. It doesn't have the beauty and simplicity of mathematical fractals, but these is like a mind meld of mathematics simulating brain neurons to recognise images. It turns out, these artificial brains, work like a dream they can recognise speech, images, hand writing and pretty much anything they are trained. This is a new dawn of AI and we are seeing the insides of a brain how it's recognising objects. So these are "organic" artificial brain "fractals". Their similarity of psychedelic experiences is definitely no coincidence. They simulate how we think the brain works.
|
|
|
Logged
|
|
|
|
cKleinhuis
|
|
« Reply #6 on: July 06, 2015, 11:32:00 AM » |
|
fractals are everywhere, each human being is fractal, everything is fractal, the feedback loops that make up the ai IS the underlying concept, your brain does use feedback loops to strengthen pather or something, same here, so definitely fractal, the thing is a chaotic process occurs as soon as analyzing the logistic map function, it is a fractal, putting thousands of feedback loops together, train them with images, nevertheless, the main thing here it is using fractal techniques all over the place, get rid of the thinking that just a for a..x loop that generates mandelbrots is fractal
|
|
|
Logged
|
---
divide and conquer - iterate and rule - chaos is No random!
|
|
|
Sockratease
|
|
« Reply #7 on: July 06, 2015, 12:14:56 PM » |
|
fractals are everywhere, each human being is fractal, everything is fractal, ... Why, if EVERYTHING is a Fractal, do I constantly get told that things like this are NOT Fractal? Either Not Everything is Fractal, or my image manipulations in Fractal Generators Are Fractal Too!! [/sarcasm]
|
|
|
Logged
|
Life is complex - It has real and imaginary components. The All New Fractal Forums is now in Public Beta Testing! Visit FractalForums.org and check it out!
|
|
|
kram1032
|
|
« Reply #8 on: July 06, 2015, 01:04:51 PM » |
|
flexiverse: Nobody knows exactly how it works, but mathematically it's actually fairly straight forward. Just a bunch of vectors. The network learns vector representations for words given to it based on example images. It's a high-dimensional vector space but that's all it's doing. Through iterative learning on more and more examples, the network attracted to solutions which are similar to the pictures. And to produce those images you basically iteratively ask the network what it sees and enhance it. Eventually, the image converges. - It has found an attractor. This aspect isn't what isn't exactly know. What IS hard to know is the precise encoding of patterns, since the involved vectors become more and more complicated. Basically, we don't know the precise formula but we do know the general process.
|
|
|
Logged
|
|
|
|
eiffie
Guest
|
|
« Reply #9 on: July 06, 2015, 01:51:18 PM » |
|
If this is a new art form then WHO is the artist? It seems you have to give some credit to the AI although even at this early stage you can somewhat guess the output for a given input image. You would need to credit the trainers as well... although I think someone at Google got bored and ran Dogs Playing Poker thru it for a week.
|
|
|
Logged
|
|
|
|
kram1032
|
|
« Reply #10 on: July 06, 2015, 03:25:58 PM » |
|
I personally wouldn't yet call the images of this AI an artform. This is more like a show-and-tell. You ask it what it sees and it describes that to you. However, this AI could be expanded into one which can decide on a theme and then render a picture based on its knowledge, to best possibly convey that theme. That, then, would be closer to actual art in that, in a way, it involved deciding on an abstract notion which then would be worked out in a creative manner.
|
|
|
Logged
|
|
|
|
Chillheimer
|
|
« Reply #11 on: July 11, 2015, 01:00:37 PM » |
|
Why, if EVERYTHING is a Fractal, do I constantly get told that things like this are NOT Fractal?
because those who tell you haven't seen the light yet I personally wouldn't yet call the images of this AI an artform. This is more like a show-and-tell. You ask it what it sees and it describes that to you.
I don't know either if you can call it a seperate artform. where do you draw the line? why put everything in labelled boxes? I'd say that this is a tool. a very advanced tool that drastically changes what you put into it.. and it is relatively easy to achieve cool results. but so do other art-techniques. and i find that it's already becoming obvious that the majority of pictures all "look the same" and are not special, not art, just trying a tool. but some are different. it starts getting interesting when you combine the different layers and change the parameters to actively form the image.. then again, those images that are generated from just noise... it is art. because it evokes emotions in me and i really enjoy looking at it. and the artist is... "conciousness"? or knowledge? AI seems so undefined... strange realms we're heading into..
|
|
« Last Edit: July 11, 2015, 01:06:42 PM by Chillheimer »
|
Logged
|
--- Fractals - add some Chaos to your life and put the world in order. ---
|
|
|
kram1032
|
|
« Reply #12 on: July 17, 2015, 09:24:08 PM » |
|
The concept of labeled boxes is a rather important one, fundamental to our reasoning and also to maths. (That's basically what category theory is all about) However, identifying what, exactly, is a box, as well as what's inside a box, can be a really messy subject. Often, multiple seemingly contradictory views of what's a box and what's in a box and what's the relationship between different boxes, can actually all be valid. If we have found a box then that allows us to talk about it. That's why we so desperately try to see them in everything and everywhere. For instance, a tool is a box containing all the actions that can be performed with that tool, and it itself is inside a box -- a tool box. (Or perhaps a tool shelf but that's virtually the same thing) As such this is a box and it's inside another box which may be in another box (a tool shelf probably is in some kind of workshop or garage, for instance) The very notion of such boxes within boxes, and relationships between all those boxes, is, on a mathematical level, how everything is made up. And it's also a very fractal notion. I think AI is a lot less undefined than consciousness.
|
|
|
Logged
|
|
|
|
|