So Google just released their in house deep learning framework and everybody in the deep learning community is freaking out.
http://tensorflow.org/
and on github:
https://github.com/tensorflow/tensorflow
>probably the most important thing in technology right now
>nobody on /g/ gives a fuck
CLOCKS ARE TECHNOLOGY AMIRITE GUYS
What's deep learning?
>>51279461
This board isn't about real technology.
>>51279242
Can I make my virtual waifu now?
>>51279549
yeah this board is for extreme consummerism and shilling
>>51279242
so can my computer now learn to rice its desktop better?
>>51279559
That's what I'm planning on
>>51279242
I bet this thread will archive with <40 replies
>>51279903
meanwhile
>RATE MY LINUX DISTRO DESKTOP
>823 replies and 437 images omitted. Click here to view
I sometime think there are only a few real users on /g/, and everyone else is just a bot or CAIMEO. Google collects sytem data and screenshots for ricing and gut and battlestation threads, while the bots poke memes at each other endlessly in other threads. Its a fucking nightmare.
/pol/ is full of paid bernie and trump and hillary shills, but at least thay discuss shit. Everything goes under the radar here.
>>51279939
Nah, its 90% shills and shitposters.
>>51279936
That's not actually what happens in those threads, they're actually blogs and chatrooms at this point.
>>51279991
1 go in it battlestation threads to see if street guru has killed himself yet
Every time, its a disappointment
>>51279242
>everybody in the deep learning community is freaking out.
Explain
i don't think anyone knowd what deep learning is
>deep botnet
>>51276714
Google what are you up to?
>No feminine penis
>No rice
discarded
>>51280062
Make money with/while making the world a better place to make money in. Skynet entered a recursive loop.
>>51280072
>feminine penis
I get really fucking mad seeing this thrown around so often. What the fuck are you actually referring to? A cleanly shaven dick on a transexual? A small 'cute' dick on a transexual? A well groomed trap goating you?
If you like dick at all, gay or straight, why would anybody every want a floppy hairy small dick? Everybody wants a clean, aesthetic large dick.
>>51279242
and
>>51276714
are having sex.
both reference each other in their posts.
>>51280191
>goating
What's this mean?
>Everybody wants a clean, aesthetic large dick.
That's all subjective my man, best not to divulge into it here as it's not really relevant to the board or the thread. Even though it would keep the thread bumped it would be disappointing to the OP, please be considerate.
Machine learning is a joke. The real way to AI is LISP.
>>51280191
What you need to understand is that the software that google just released can find the best dick.
You can find an answer to this question. You can get the software to identify dicks and then rate them.
I'm 100% serious.
>>51280293
>What's
Wait does that mean "what is", I use that vocally as "what does". I might be an idiot.
>>51280317
I'll make the logo
>>51280295
Elaborate.
>>51280293
Goating somebody is when you put your entire Johnson and friends behind your thighs and close your legs. Pretty popular in trap treads. I'm sure they don't call it goating though. One dude doing it to another just to fuck with him is when its called goating.
>>51279242
> TensorFlow was originally developed by researchers and engineers working on the Google Brain team within Google's Machine Intelligence research organization for the purposes of conducting machine learning and deep neural networks research.
>TensorFlow is an open source software library for numerical computation using data flow graphs
Got my hopes up for nothing. This is just a math API. Literally thousands of them everywhere.
>>51280377
It's just a math API in the same way that every single computer program ever made is just a couple of maths calculations.
Calling software 'just maths' is like calling plastics 'just hydrocarbons'.
>>51279242
>The TensorFlow Python API requires Python 2.7.
fucking dropped
>>51280434
OP said
>Google just released their in house deep learning framework
That implies that I can somehow use this to make a deep learning application without spending 50 hours programming it
>>51280435
Fuck you, Guido! Python 3 is shit!
I wouldn't know how to use this
and even if I did, I wouldn't have any use for it.
It's nice that it's open source; thats about all I can say.
>>51280559
50 hours is not that much time to create a deep learning application.
>>51280323
You are, but not for the reason you think.
>>51279242
>hurr durr deep learning
fuck off with your buzzwords
>>51280592
>implying I don't have a job
fucking neetfag
>>51280368
Like that guy from "The Silence of the Lambs"? Interesting.
So to goat someone is to tuck your willy in and show it off to them like mooning but from the front?
>>51280618
>on /g/ at 10:30am on a Tuesday
>I h-have a real job
>>51280618
That's not even an argument. Unless you were specifically talking about you as the person.
Yes I agree, YOU wouldn't make a deep learning application.
However, someone that gets paid more than minimum wage will.
WHAT is this and why should I care?
>>51280653
>spot the ameritard
>>51280618
not him, but 50 hours is probably only like 2 weeks worth of work, assuming you only spent 12 hours per off day doing it.
>>51280609
>>51279461
no one knows what this is even avout, heck i dont
>>51280653
Your neet is showing desu
>>51280663
its machine learning software
the program can learn and grow on its own, depending on the data you serve it. Remember pic-related? Yeah, that was made using this software
this waifu upscaler also uses neural networks(not google's)
http://waifu2x.udp.jp/
>>51280724
i dont get these images, what is the meaning behind them / the way theyre generated?
>>51279242
Aside from the CLA and apache 2.0, it's basically mxnet or theano++. It's slower than mxnet for now, though, so people aren't going to all converge on it just yet.
>>51280368
Yup. Forgot that movie existed. But yeah that's the gist of it. Don't see the appeal or the funny factor though. I definitely don't want to see any of my friends dick n balls or their assholes. Probably don't wipe that good anyways.
>>51280751
read this first
https://en.wikipedia.org/wiki/Artificial_neural_network
>>51280751
it's so ai can recongize images by looking at them visually.(no hashes, pure noise data)
>The ultimate goal is to have a program that could see any kind of picture with a dog in it, for example, no matter the lighting, orientation, color, etc., and tell us there’s a dog in the photo.
http://nerdist.com/why-are-googles-neural-networks-making-these-brain-melting-images/
>America wins again
>People doubted America
this is more science than technology
>>51280029
It's the only framework that has EVERYTHING (symbolic computation, speed, multi-gpu training, parallelism, cudnn support, all the non-linearities, and training algorithms you'll ever need). It's also the only framework actually built by engineers as opposed to scientists who know fuckall about programming as is usually the case. By this time next year, it's probably going to be THE framework everyone uses.
>>51280435
Nobody uses python 3, it's the correct choice to make here.
>>51280072
Kill urself tripfag
> ;^)
>Great piece of technology gets released
>lolol it uses python..!
>anime reaction images
>botnet
>lool kill urself
>yay murica
I see why no one over the age of 21 wants to browse this place. It's literally for fucking children. Can't believe I didn't see it sooner.
>>51280953
Nice contribution.
>>51280953
we just don't know the potential this thing has yet
>>51280822
>>51280822
Fuck off this is more about computing than it is science
>>51280559
Lol what did you think it was? You just speak to the computer and it does what you want?
It gives you a visual representation of the data flow to tinker with, it doesn't get easier than that. I'm sorry that math is hard for you but how do you think you're going to build a machine learning program without having some rudimentary understanding of how it works?
>>51281014
>People still thinking that /g/ discuss computing
Honestly they should create a board that all the shills, rate and desktop threads could go to, and leave /g/ to actual computing stuff like this
>>51281072
Don't forget the phone threads. All belong in the
>>>/trash/
>>51281072
It wouldn't work. the 'actual /g/' board would be highly inactive.
look at 4+4chan's /tech/ board. In order to contribute significantly to topics like these, you'd need to already do this shit for a living or specialize in that field. Programming has many fields. an average joe programmer wouldn't be able to do much with it unless they specialized in machine learning.
>>51280295
This.
I hope to become a formal member of the Knights of the Lambda Calculus someday.
>>51279242
what exactly can I do with it? I have no clue about machine learning. Can you give an example where it could be applied?
>>51281119
>i want to be just like lain from muh animes
Fucking weebs.
>>51279242
>everybody in the deep learning community is freaking out
shill plz
>>51280295
Kek.
>he thinks a language is a method!
>>51281133
nothing. really.
you must have some deep knowledge about machine AI & learning.
>>51281155
He is right, though.
>usual day: 10 posts on the list
>tensorflow released
>80 posts in under an hour
>>51281133
http://waifu2x.udp.jp/
Can we modify it so it instead of dogs, seals and butterflies it finds anime grills?
>>51281069
I WAS expecting something liketf.setfitnessfunc(func)
for _ in range(5000):
tf.input(randomdata)
tf.epoch()
And then done.
>>51281198
stretched anuses
>>51280559
It takes about an hour to setup an elaborate graphical model trained by variational inference. Less than that if all you're doing is classification with CNNs.
>>51281150
your pleb is showing buddy, i was referencing that kid taro
>>51281195
with the progress made in the field you could make a neural network capable of tagging anime pics, yes
>>51281159
>method
oop in loo pajeet
>>51281215
>>51281225
Say what you want. But OP was extremely misleading when he said "released their in house deep learning framework"
It's possible to use this for deep learning. But it isn't packaged with deep learning. This is practically just a math api
>>51281175
Notice how fucking clean the waifu2x resized one is.
It resized it as if it were a fucking vector image. That's the power of MACHINE LEARNING
>>51281281
so they basically give you the formula, but it's up to you to learn how to use it
>>51281198
It's more likelayers = []
layers.append(LinearLayer(input_size=784, output_size=1000))
layers.append(ReLuLayer(input_size=layers[-1].output_size, output_size=2000))
layers.append(SoftmaxLayer(input_size=layers[-1].output_size, output_size=10))
alg = SGD(lr=1e-3, rule=ADAM())
cost = MSE()
model = Model(layers)
data = HDF5Dataset("mnist_train.h5")
for (x, y) in data:
alg.train_batch(model=model, data=x, label=y, cost=cost)
data_test = HDF5Dataset("mnist_test.h5")
print (model.fprop(data_test[0]) != data_test[1]).sum()
>>51281281
What do you think deep learning even is?
>>51281281
>It's not a pizza, it's bread with tomato sauce on top!
>>51281292
It's better than anything we have but I still don't like it. I get real autistic over "original" purity and lossy conversions. I'd rather have the original than a modified copy, I'm not even really sure why other than the loss of details sometimes. I don't trust this thing.
>>51281319
idk some sort of norwegian metal band for kids?
>>51281321
You're proving his point.
it's not the full thing.
>>51281355
it's up to you to add the cheeze and the salami
>>51281340
>I get real autistic over "original" purity and lossy conversions
I'd never recommend replacing an original of course. I just think it's nice what potential it has.
>>51281380
>salami
>pizza
absolutely disgusting
Wow I gotta play around with this once I have time
>>51279242
>http://tensorflow.org/
>video starts playing
>indian accent
Every time
>>51279242
What is "deep learning"
>>51281468
Ok.
>>51281468
Like learning with regular neural networks, just with way more layers instead of just 3.
>>51281418
Oh for sure, I didn't mean to detract from the merit of such a tool. I just get scared when things like that come around and people start replacing their originals with upscales.
"m-muh archival" -me
>>51280360
https://en.wikipedia.org/wiki/Confidence_trick
>>51280360
>>51281515
I'm sorry I meant related to LISP and AI specifically, what's the link between those two things, or rather why is LISP in particular relevant/good to/for AI.
>>51279991
Agreed. Just a daily circlejerk for tripfags
>>51281596
Lisp was created for AI, literally machines with the intelligence of a man, which was promised by some time in the 70's. It seems like nonsense today, but people actually believed this.
The government pumped money into Lisp for AI for decades but it failed to deliver. The researchers who were still serious about AI moved on to other ideas, like neural networks, hidden Markov models, and expert systems.
This failure caused the "AI winter" where companies and governments were disillusioned by Lisp. They dumped millions if not billions of dollars into Lisp and didn't get anything useful for AI. This "AI winter" is when Lisp programmers started blaming everyone else for Lisp's failure, which continues to this day.
>>51282160
Thanks for the information, I appreciate the story. I've always wanted to learn a LISP because they seem so alien, I don't see any merit in them (I'm not saying it's not there I'm saying I don't know of it) so I'd like to learn it just to have some insight on it, to actually know what it's about.
>which continues to this day.
kek
>>51282160
Expert systems were such a failure, it was hilarious!
>>51281292
so you mean we get better quality chinese cartoons if machine can learn?
>>51282201
They have the most convenient, yet powerful macro system out there. Symbols are fantastic (basically "global enum values" where the name is used for display but the "hidden numerical representation" (conceptually) is used for other operations and it doesn't compare to numbers), hamoiconicity and regularity are ridiculously nice. Those are probably the main points about lisp.
>>51280861
>python
>the correct choice to make, ever
>>51282297
tensorflow is mostly c++
>>51282297
I agree, but it is what it is; over 90% of ML people use python because it's "easy to write experiments with" (interpreter is a legit point, but mostly it's because people don't know about other languages).
>>51282319
Tensorflow is written in C++ (and CUDA), but has a python API. That's how most python tools are usually implemented: C or C++ in the backend, python in the frontend.
>>51282234
One day I will get around to it. I have this big fear that I'll spend a big amount of time learning something that will have 0 practical use to me, I've already learned a ton of things that have been obsoleted.
The only application for AI that I can think of that's even remotely related to me would be video games but I doubt I'd ever write one of those even. My lack of sleep is causing me to be verbose.
>>51282338
If you plan to work for naughty dog, lisp is a good language to learn.
Also, lisp is one of the easiest languages to learn (because of the regularity, mostly).
