Why aren't you learning machine learning so that you can make your own AI?
can't I just get a machine to learn it for me?
How do I start, OP-kun?
I've looked into some. Waifu2x for example is a generic machine learning. If waifu2x were given weight values and you taught it to optimize for the best values, then it would produce even better optimizations.
I have some ideas on what type of programs to test on. You could try doing a alphago type bot for example.
>>53423937
Except I'm literally taking machine learning this semester.
Also my professor is a fucking madman who spends 25 hours a day tutoring on the side, and answering stackoverflow questions
>>53424181
>Prof wastes time on SO
You must go to shit college.
>>53424302
It's pretty shitty, but this is the first year this guy is a professor, and he's basically fresh out of uni himself.
Really tries though. Records all his lectures, posts summaries of class and shit.
Literally works his ass for the class
>>53424302
>Professor does lots of teaching on the side and know what kind of knowledge usually is lacking in the industry.
>considered a bad thing.
Yeah, professors who don't have a clue and who doesn't care about teaching is much better.
>>53424143
Start by understanding evolution and the scientific method. Then you can move on to graph theory, linear algebra, generalized linear algebra, combinatorics, information theory, statistical dynamics, criticality, and Newton's method. Then you can start creating really stupid AI.
Then learn about the more advanced forms of gradient descent involving momentum and Taylor approximations, mean field theory, various kinds of regularization, gradient boosting techniques, bagging, and boosting.
The combine all of the above with reinforcement learning, and you have AlphaGo.
>>53425634
Also, it doesn't hurt to understand general topology and representation theory. All things said, machine learning really isn't that complicated.
>>53425634
>>53425669
Don't be a dick, anon.
remember that time /g/ went crazy with AI evolution?
>>53425634
I'll be dead before I get to half of it. Might as well do nothing.
>>53425634
>>53425669
Does programming come into it somewhere?
>>53425723
He's deliberately pulling in areas that practically, you don't need for machine learning. I think it's for him to feel cool on an anonymous imageboard.
Just my opinion, but you could also start with this:
https://medium.com/@ageitgey/machine-learning-is-fun-80ea3ec3c471#.pjn70vsxd
I also recommend the Andrew Ng course on Coursera. It uses Octave but the language shouldn't matter much, anyhow you could just apply the lessons to any project you want.
>>53423937
>muh singularity
Please fuck off, singularities are as dumb as time machines. All machine learning AI does is multiply and add matrices together.
>>53425877
>Andrew Ng course on Coursera
It's not a very good course.
>>53425685
Fine. Learn statistics, linear algebra, and gradient descent, then screw around with the equations.
If you actually want to understand why these things work, you really do need all of the above, and that's not even a complete list given our current inability to create human-level intelligence. You're trying to automate intelligence. What were you expecting?
>>53425723
Only if you want to experiment with things people haven't done before. There are tools that let you just mess around with stuff at a high level, like Mathematica.
>>53425719
Understanding the theory is a consequence of being interested, playing around, and being willing to learn. 27 months ago, I knew almost none of the things I mentioned. I spent an ungodly amount of time playing around with the stuff, reading math and AI blogs, reading wikipedia, watching AI youtube videos, pacing back-and-forth thinking up crazy theories of why this stuff worked and how to make it work better, and testing my ideas. This was all while balancing a full-time job that's completely unrelated.
It's worth it. It's absolutely, 100% never-let-anyone-tell-you-otherwise worth it. Once you get into the flow, it's really fun, it's intuitive, and it takes you to the deepest regions of math and physics. It also builds your intuition like nothing else. It makes you think fast, and it gives you a huge number of tools to make connections and reason about things you might have thought were guesswork.
I started by watching youtube videos of Geoffrey Hinton, then playing around with python and theano based on the code here:
http://deeplearning.net/tutorial/rbm.html
If you're on the fence, just do it. You can thank me later.
>>53423937
Because it's just a gimmick at this point unless you are using deep neural networks and stuff that no one understands but somehow works.
Statistical approach achieves better results than regular networks most of the time with less overhead.
>>53425877
All of those things are directly related to machine learning. The first line covers the fundamentals on which deep learning became successful. The second line covers the techniques people applied to deep neural networks to improve them. The third line was literally the missing piece that allowed deep learning methods to be successfully applied to Go.
>>53425984
>it's gimmick
dat denial
>>53423937
I've been learning how to use tensorflow instead of sleeping all week. I don't want to miss that bandwagon.
>>53424302
How could it be? People literally take courses there.
>>53425926
Philosopher here, gotta presume ure European or American
I am not american nor european, describe in few words who you guys think is the mind that is going to prove that Strong AI is actually possible in some way?
Been looking at Fodor for a while but he seems like a bitch to being concrete about his theories. Chalmers went a bit off from more actual problems, Penrouze just said his word and not gotta prove shit, same with Searle ( I am putting it all in simple terms ofc). So, who is that guy or it is all the same on ur board too?
>>53426540
>Philosopher here
I want my burger medium rare
>>53423937
>Why aren't you learning machine learning so that you can make your own AI?
no thanks
>>53425984
Mostly true. People should probably take a computational statistics class from the stats department first before taking machine learning offered by CS faculties.
>>53426425
You know you are in denial when memes are your only arguments.
I like reading in statistics communities, how bum ravaged they are that all these rad words like Deep Learning and Machine Learning and Data Science are getting buzz, when they're all basically just applied statistics.
>>53426901
they're just mad they're gonna be on the street in 5 to 10 years.
>>53424143
Search for Andrew Ng's machine learning course on coursera. It's still the very best introduction to machine learning, and you don't really need a lot of prior knowledge.
>>53427011
I've always been recommending that too but I have to say http://www-bcf.usc.edu/~gareth/ISL/ these people are pretty good too. There's a MOOC through stanford, and that link has two texts, one more in-depth than the other. I've only just started it so I can't compare it to Ng's fairly, but hey it's free.
>>53424143
>>53427011
This
>neuralnetworksanddeeplearning.com
Is also okay if you're only looking for neural networks, seriously lacks background though
>Murphy's MLAPP
>Bishop's book
I liked murphy's best. These books focus a lot on data science though
>Norvig&Russel
Best AI imo, though a bit old
>>53423937
If I'm being honest, it seems like it must be pretty challenging. I can't say I wouldn't enjoy knowing some, though.
>>53427090
Depends how deeply you want to go into it. You really don't need more than partial derivatives and some linear algebra to do something basic, and you could probably get away without the linear algebra. And really, you could get away without either by using libraries.
i've read hundreds of papers on backprop
it's all a fucking meme
>>53427114
I'll give it a shot. Gotta go find a good book first, I guess.
>>53424181
>>53424326
I had an instructor (he taught the main lecture and the lab) like that once, but he was finishing his math PhD at the same time. He must have evolved past the point of sleeping or something. Professors who work like this are bro-tier.
>>53423937
ML is boring. Too much statistics and probability theory.
>>53423937
Because ML is boring and half its application is building botnet tier stuff.
If I had the time, I'd rather learn more about compiler construction, audio processing and parts of computer vision that can be done with simple edge detection.
will EE have a place in our new AI world?
>>53427782
> Implying AI would run on something other than semiconductors.
I dropped the class
But OP, I'm taking masters degree about it
Super informative thread. Thanks for the resources, anons.
>>53425634
>>53425926
>>53427050
