Has something like this been done before? >train neural

DUDE WEED LMAO

>>57538734
???

>>57538712
For what purpose, exactly? At best, you'd get a bunch of false positives and it'd take a *really long time* for anything remotely "real" looking to be generated randomly if that's what you're after. This would cost a lot of money for basically no reward, so everybody loses, and who wants that?

>>57538772
With enough training data you can minimize false positives. With proper gpu parallelization you can generate images very quickly.

Think about it anon. It *could* generate an image with blueprints for an FTL engine. Or maybe Hillary Clinton getting fucked by niggers. If you ran it for long enough just imagine what you could find!

>>57538772
Generate cheese pizza m8

>>57538808
I think you seriously underestimate the time it would take for even *one* kinda-sorta "realistic" to be generated randomly. Hint: it'd be much faster to contruct an FTL drive the old fashioned way and even faster to pay a bunch of niggers to break into Clinton's house and rape her. Hell, they might even do it for free.

>>57538861
I understand how long it would take, but I'm still interested in trying. All I want to know is if it's been done before..

>>57538808
>>57538861
If it was a black/white pixel image of 100x100 it would need 2^10000 images to iterate all of them
2^1000 is roughly 10^301 aka 10 followed by 301 zeroes.
The extimate number of atoms in the universe is 10^98
That's still 2^1000 and not 2^10000, so you take that 10^301 and elevate it to the power of 10 to get how many images you need to generate

>>57538808
do you have a high school diploma?

>>57538712
>sax plays
>WOKE UP THIS MORNING

>>57538712
/g/ has gone over this before. It's not so much determining what might be "real" or not it's the sheer volume of data required to make the project viable.

If you set this up with some national dick waving tier supercomputer then set the image filesize to 50kb and waited, you'd be dead before you got any real result. Your kids would be dead. Their kids would be dead. Our brains can't even register how many possible binary combinations are in 50kb.

>>57538897
That's to iterate all possible images. What's the ratio of noisy to realistic images? It would take far less than that before you got a realistic image.

https://youtu.be/5Yy_unGaD-w

And this is only for 256x256 greyscale images.

>>57538772
>For what purpose, exactly?
Legal CP

>>57538879
>I understand how long it would take
No, you really, REALLY don't.

>>57539092
>Legal CP
Just fucking draw the CP or use 3D models/animations. People are doing that *right now*.

>>57539092
>Legal CP
t. coward who's afraid to write CHILD PORNOGRAPHY.

>>57539108
>tfw no algorithm to draw it for you

>>57538712
No one has ever thought of this before!

>>57538879
It would take a cluster of computers a year to generate a single sensible image, and only if you limit it's resolution and colours. So it's impossible to get proper FTL blueprints even in a million years.
>>57539131
Reported

>>57539070
>What's the ratio of noisy to realistic images
impossible to tell, as you would need to generate a fuckton of images and sort them by hand before you can even calculate the ratio, which kinda defeats the purpose.
You can't train a machine to think like a human unless you feed them human-processed data beforehand, at which point you don't need a machine doing it for you

It's much more effective to get it painted in China.

>>57539231
That's the real deciding factor on plausibility here, NOT the number of possibilities an image can have. You can train a neural network with a relatively small sample (~10k maybe?) and have it do the rest. If it does find some "realistic" images, you can estimate the ratio.

Anyways I don't pay for electricity so I think it would be interesting to just run this shit on my computer for a few weeks and see what happens.

>>57538712
Someone did that with Yahoo's open source nsfw detection AI thingy
https://open_nsfw.gitlab.io/

Weirdest boner ever

File: 1450619494714.jpg (35KB, 572x380px) Image search: [Google]

35KB, 572x380px

>>57538712
Try

/dev/urandom | grep Apple

And you'll understand.

Oh look, it's this fucking thread again

>>57538712
Google are literally doing it right now.

>>57539392
>a few weeks
Nigga, if you went through 1 BILLION images per second it would take 6.33^2993 YEARS to iterate through them all.

>>57539413
Neat, thanks anon. Looks like the images they found are too abstract to be used in any meaningful way, though. Guess I won't do it.

>>57539450
sauce?

File: 1458082312239.jpg (28KB, 365x337px) Image search: [Google]

28KB, 365x337px

>>57539440
this

>>57539461
https://aiexperiments.withgoogle.com/

>>57539457
Again there's no need to iterate through them all just to find a couple pictures. I don't know why you keep bringing this up.

>>57539509
You really aren't appreciating the enormous amount of time it would take to find anything. I'm talking greater than the age of the universe here. A few weeks is a drop of piss in the ocean, and I guarantee you wouldn't find anything. The majority of the images would look like a random mess of pixels, and you would need some neural net to sift through all those. 1 billion per second is a huge exaggeration on how fast it would be able to do that.

And this is just talking about a 100x100 image with 2 colors.

>>57539131
M-MODS

>>57539509
I'd type the percent of images you could go through in a few weeks, but the amount of zeroes in the decimal is way over the character limit of 2000.

>>57539569
>>57539169
It's not against the rules to call disgusting mental illnesses by their name.