Cutting-edge projects in AI

If CS failed to simulate a scientific field how are they supposed to simulate a brain.

Generative modeling is the most exciting thing right now

>>8697765
Links/resources? Sounds interesting...

>>8697712
Solving starcraft, it's a partially observable domain.

If we add enough layers, can deep learning solve this problem too?

>>8697782
https://youtu.be/rAbhypxs1qQ

this stuff is pretty insane. Give the network a sentence, it produces an image. This whole field has basically exploded in just the last year, thanks to a thing called "Generative Adversarial Networks"

>>8697747
fpbp

>>8698097
Holy shit

>>8697747
kek

>>8698097
>However, I am sure that in 2017, researchers will start working on generating full HD animations with an improved version of this architecture

At last, now I can start directing my own 2D hentai

>>8698097
Will look at this sounds very cool.

>>8697712
sparse distributed representations

>>8699300
I'll google it , but do you have suggested links on the topic?

File: canvas.png (867KB, 1600x800px) Image search: [Google]

867KB, 1600x800px

>>8698097
Not as impressive, but still quite nice :
http://affinelayer.com/pixsrv/index.html

File: Joseph_Wright_of_Derby_The_Alchemist.jpg (535KB, 600x787px) Image search: [Google]

535KB, 600x787px

when will you realize that CS is the alchemy of the 21st century and true AI is its philosophers stone?

>>8699614
>http://affinelayer.com/pixsrv/index.html
fuck,that's mad cool! Honestly this technology looks so futuristic....

>>8698097
>>8698432
Exactly my thoughts
>implying this wont instantly be used for porn the minute after the release of the source code

File: 220px-Douglas_Hofstadter,_Stanford_2006_(crop)[1].jpg (12KB, 220x223px) Image search: [Google]

12KB, 220x223px

>>8699634
>tfw you understand this Analogy

>>8697712
The most exiting stuff is all behind closed doors.

>>8698057
I think solving starcraft (I'm talking about brood war, but I guess the same goes for SC2) is a bit far fetched.

However, I think it would be fairly straight forward to program AI's that could beat humans even with suboptimal play just because of how insanely good their micro would be. Hell, I doubt deep learning and neural networks is even necessary for this.

File: 1b9.jpg (43KB, 640x539px) Image search: [Google]

43KB, 640x539px

>>8699779
>tfw the second islamic golden age will give rise to ibn Hayyan 2.0

>>8699792
>islam
gas yourself

>>8699784
Most interesting companies to look at? Is it mainly the "Big Guys" (Google, IBM, Microsoft, Facebook) or is it there some interesting startup?

>>8699861
>software

It's also heavily about hardware. Nvidia, Google, and closed door are most important. Startups with a few months of work surpassed Google's self-driving technology using latest hardware. A lot of the techniques are unimportant compared to small fundamental changes in the methods and better hardware.

>>8699894
That's a reasonable point , even if I didn't think hardware would be as important as you make it sound. We'll see.

Best AI project was Accelspammer.

File: 00000.jpg (27KB, 768x432px) Image search: [Google]

27KB, 768x432px

>>8697747
keked

>>8699752
>porn
it's been done
https://open_nsfw.gitlab.io/

>>8700266
It learned how to recolonize penises LMAO

>>8701599
recognize*

>>8698097
That's cool as fuck and that whole channel seems pretty great thanks

>>8697712
Do you guys think we'll manage to build a sentient machine in like 300 years? What are the major challenges?

>>8697712
Every time you fill out the picture captchas you are helping googles deepmind projects learn how to destroy the human race, and obey traffic laws while doing so.

>>8699779
Dont think you are smart for understanding it. Its brainlet tier popsci. Not to mention, theres no reason why strong AI cant work (unlike the philosophers stone). With complex enough computers, networks, and software we will one day be able to simulate a human brain down to the atomic level in real time. Who knows if it will ever gain consciousness, but thats not really the point.

>>8697712
Genetic code into C programming.
Simulated with environmental variables at break neck speeds. This is the future anon.

Nothing because it's all a bunch of bunk that's more reliant on graphical power than advancement in theory. It's been stuck at graph/neural node models for decades. At this point all that's been happening is people making bigger networks with more nodes on more expensive machines with more graphical power. We are at the end of the line in the field as far as theory goes and tangible material advancements will come along as stronger hardware is developed.

We'll see strong AI when we get NVIDIA 7Billion or some shit.

>>8701848
well genetic coding is cool, but calling this the future is quite funny

>>8701843
Mate chill, it was more a joke on analogies and Hofstadter...

Does anyone know of there's any work that has been done regarding establishing hardware in a neural net fashion? My understanding is that we are simulating nets on computer architecture and I think it would be cool to see how a net could function on hardware specifically designed in the node and layer fashion (instead of pieces of a computer chip) and maybe the weights could be scalable settings on each individual node on the machine.

>>8702283
I cant really give an informative answer but I heard intel is working on machine learning specialized hardware which has better performance by factor 100 and more

>>8702372
Dang that's cool, it makes sense to be fair

>>8699614
>http://affinelayer.com/pixsrv/index.html
Btw pewdiepie just referenced this link on his latest video...does he browse /sci/ ??

>>8702372
That sounds pretty neat. I wonder how much it would cost to buy some or if they'd even sell it commercially.