Why isn't anybody using deep learning to build Emulators?

Get on it then, faget.

lol d00d just imagine broooo

>>62096949
Pretty sure OP is one of those people without even basic understanding of programming who constantly bothers his programmer friends with all those great ideas for projects they should make...

>>62096920
cool , now how do we train it
>just insert iso
mfw 40gb training data

sounds pretty interesting, but I have no clue what any of that means from a technical standpoint desu
maybe someone is/could do it if they knew what do

I'll make the logo

File: high.jpg (170KB, 800x800px) Image search: [Google]

170KB, 800x800px

>>62096920
how high do you even have to be to think that's a viable approach?

>>62096920
oy vey shut it down!

>>62096920
Idk, training data is too small. There aren't too many games as it is, and less that are both on console and PC. That's the only way you're getting training data (besides taking emulated games).

>>62096920
Neural network aren't magic and aren't 100% accurate. You need 100% accuracy for your ISO : 1 bit different => corrupted ISO

Neural network is one of the buzzwords, like "blockchain" for example : it's hype so lot of people who have no experience in engineering nor mathematics start to think that these "magical computer programs" can make whatever they want.

I have those kind of people at work, it's hard...

Of all the neural network/machine learning projects and potential use of the tech, this nigga wants to make emulators. This is like when they had a fucking hologram of 2Pac.

>>62096920
>So you load the COD PS4 ISO and the COD PC ISO.

Surely they both have EXACTLY the same CONTENT but with a different disposition right?
i mean how hard could it be to train a fucking AI to learn to reverse engineer a fucking encrypted obfuscated undocumented machine code , convert all the assets,rewrite it all from a specific environment with specific hardware to a very diverse environment with very diverse hardware..

cool story OP , totally feasible, can't wait to donate shekels monthly for this project
I fucking hate this place more and more everyday.

Go home, you're drunk. Go home, you

are

drunk you motherfucker.

File: oak_ridge_titan_supercomputer.jpg (84KB, 1200x794px) Image search: [Google]

84KB, 1200x794px

>>62096920
"And here is where I load my ISO when I want to play a game"

Sounds like a multi-armed bandit problem to me and that's np-complete

>>62096920
I don't think you understand how neural networks work.

Guys what if we used a neural network to transform my internet into 100 gigabit? It would learn how to trace the packets and move them more efficiently through the pipes

>>62098442
Or it could learn what content you download and just generate it on your computer before you even look for it.

File: 1476465942991.png (1MB, 1150x752px) Image search: [Google]

1MB, 1150x752px

>>62097158
This

Why not just use deep learning to make perfect OSes

Just markov chain them.

I don't think the difficulty lies in recompiling the binaries, I think it comes with implementing the interfaces with which the software on top interfaces with.

>>62098726
Is this a program that asks you preferences and spits out images of anime girls that fit the criteria? Do you have a link?

>>62098813
Why not just use deep learning to make perfect sex robots?

>>62098726
>>62099102
I didn't want to Google it at first but I did anyway.
https://hiroshiba.github.io/girl_friend_factory/
Now Google knows.

>>62099102
>>62099177
If you want one that looks better here is a newer one based on the same concept
http://make.girls.moe/#/

>>62096920
Human beings have to explicitly set the success conditions. If you already know what a working PC version of a PS4 game would look like, then you would already have a working PC version.

>>62097012
40gb training data is miniscule. There are lots of neural maps that started with 40tb of training data.

File: be7.jpg (54KB, 500x531px) Image search: [Google]

54KB, 500x531px

>>62096920
Why don't you just get ron paul to do it?

>>62096920
wont work: not enough trainings data.
and
rnn being touring complete doesnt mean shit.
now go back to your cs 101 lessons about nn

>>62096920
"auto proters" already exist, sort of
they aren't very good, save some time I guess

>>62096920
>1. input ps4 iso
>2. magic happens
>3. output pc iso
kek thats not a fucking use case computers are deterministic so is the ps4.
neural nets are supposed to find patterns which are hard to describe in code.
I don't think it's possible and even if it was it would be horribly inefficient.

Also on what data is your neural network supposed to train binaries?
wishful thinking nigga

>>62099609
40GB is one sample in that case those 40GB sets you are talking about contain millions of samples

So, what about doing this:

1) Get the binaries from both the PS4 and the PC ISO.

2) Decompile those files back to C++ source code again.

3) Use neural network on those source codes.


Then use the following steps:

Get Uncharted ISO, get the binaries from it, decompile the Uncharted program to source code, use the neural network, generate PC uncharted source code. Compile it and profit.

>>62101586
You still don't understand how immensely complicated this is.

Would it work? depends.

But it's not going to work with a single game.

And you're also assuming that the PC version and the PS4 version are EXACTLY the same other than being ports. In a lot of games this isn't the case.

If it were to happen though, the easiest would likely be xbox > windows. They both run DirectX.

All of what I said is completely hypothetical and should not be taken seriously. What should be taken seriously though is that this is nearly impossible due to the complexity of the task. This isn't teaching an AI the difference between an apple and an orange. This is teaching an AI to port a game which would take humans a load of time (and they have the actual source code to look at, btw. decompilation may yield code but it's not the same. Function and variable names are lost, or even discarded, when a program is decompiled)

I spent more time on this post than I should have.

>>62096920
You're the guy from a few days ago who wanted to simulate materials atom by atom?