Who considers ML/ Statistical Learning a huge meme?

I think it attracts too many retards that don't realize the math required to push the field forward.

Other urls found in this thread:

projects.iq.harvard.edu/stat110/home
neuralnetworksanddeeplearning.com/chap1.html
twitter.com/NSFWRedditGif

Yeah, shit is sad desu.

Retard here trying to learn math to get into this.

Yeah, not keen on how big it's becoming. It's interesting, but you know a bunch of people are doing it for money, or because it's the "next big thing", and I really wish they'd fuck off back to software development.

I have no problem with software development. I am a programmer too. I have a problem with brainlets using blackbox libraries (which they don't understand) to solve problems and then claim they are a "data scientist". I deeply value mathematical rigor (as a math degree holder) and wish those people would just drop the field.

These sort of people would immediately drop any real graduate course on ML that required original research.

I'm a retard brainlet programmer and I've been getting by programming without any math for years, so I've forgotten all my math from school. I've taken a interest in statistical learning because I think its very interesting and it's inspiring me to learn math so I can understand this and the techniques and their assumptions and limitations. But since I've forgotten so much math I have to pretty much have to go back and start at the basics.
I guess if I work through textbooks for Algebra 1 & 2, College Algebra, Calculus, Linear Algebra, Discrete Math, Statistics I'd have the prerequisites to really start getting into understanding statistical learning. Does this sound like the right prereqs and order to learn them in?

I admire your willingness to learn.

I would learn Algebra , Calculus, Statistics, Linear Algebra, Probability theory.

Calculus, applied linear algebra, statistics. Thats what you need to do ML

Is it better to go through a Probability textbook first before a Statistics one?
I've seen in the Preface to several Statistics textbooks and Probability textbooks and Linear Algebra textbooks that Calculus is a prerequisite. Some Probability and Statistics textbooks say their prerequisites is only highschool math and I'll probably work through them as a "baby's first" textbook before going onto the more serious ones.

I'll have to do some more baby steps of highschool Algebra 1 & 2 first before I even get to that.

I'm familiar with LG so finding plenty of textbooks is no problem.

can you fag help me :3

Why is Statistical learning a meme? I agree that ML and Data science being a meme though.

>I would learn Algebra ,
Just to be sure, do you mean Algebra as in like highschool Algebra 1 & 2, and "College Algebra", or as in like groups, rings, and fields?

>math required to push the field forward.
The only reason why the field is taking off right now is because of the hardware. All the math you need was figured out in the 60s--intelligence is really that simple.

running pre made packages and contributing to a field r a little different eh

not true. likelihoods in stats for higher dimensions are newish

Most intro stats books are going to introduce probability before statistics, because a knowledge of basic probability theory (distributions, parameters) is required to understand formal statistical inference.

I wouldn't waste time relearning calculus...maybe try taking a couple of derivatives and integrals, but you'd get enough practice with that working probability problems so it shouldn't be an issue honestly. I'd only worry if you were taking a class for credit or something.

Most efficient use of your time in my opinion? Read the first 7 sections in the Halmos set theory book (very short, maybe 30 pages, google for PDF), then get a basic stats book with an intro to probability. Understand probability distributions, sampling distributions, basic estimation procedures, and hypothesis testing (for instance, intuition behind likelihood ratio test), and you should be good to go.

Could easily be done in a few months if you don't care about being super rigorous. Linear algebra can probably wait a bit until you have the stats basics straightened out.

Most intro stats books are going to introduce probability before statistics, because a knowledge of basic probability theory (distributions, parameters) is required to understand formal statistical inference.

I wouldn't waste time relearning calculus...maybe try taking a couple of derivatives and integrals, but you'd get enough practice with that working probability problems so it shouldn't be an issue honestly. I'd only worry if you were taking a class for credit or something.

Most efficient use of your time in my opinion? Read the first 7 sections in the Halmos set theory book (very short, maybe 30 pages, google for PDF), then get a basic stats book with an intro to probability. Understand probability distributions, sampling distributions, basic estimation procedures, and hypothesis testing (for instance, intuition behind likelihood ratio test), and you should be good to go.

Could easily be done in a few months if you don't care about being super rigorous. Linear algebra can probably wait a bit until you have the stats basics straightened out.

This. CNNs are the hot shit that everybody is getting excited about and the papers that underpin that are all 30+ years old.

but CNN is fake news

I'm really interested in using Riemannian geometry/algebraic topology to motivate some learning algorithms/do dimensionality reduction. But I also need to know how to be useful

I don't care about pushing the boundaries of machine learning because I'm way too brainlet for that. I still think it's possible for *regular* people to harness it for practical applications not yet invented.

Layers on Layers, user.

Im sorry if i am also a retard brainlet in your opinion, but i would be interested in your opinion of these things:
I've been into the topic of machine learning in the last time and i am also motivated into learning shit about it, but what do you think about applying the concept of using neural networks and machine learning that has proven to be succesful in for example translating texts or even simulating human speech to for example controlling robots in all kinds of fields, thus creating machines that move more naturally and can learn from their environment (i know that using an actual robot to simulate the learning process would take up way too much time, but maybe a complex realistic simulation could work out) and also when and how do you think would an artificial intelligence actually capable of solving abstract problems of all kinds by itself without specific human instruction/programming?
(excuse me if my grammar is bad, not a native, but you should get the idea)

But i have to add that i totally agree with you, that because of the way ML is presented in media (especially youtube), it became a meme with a lot of people watching videos with no actual explanation of things but instead a huge list of "possibilities" and the future and shit.

It's a massive meme, if by "meme" you mean "something that's rapidly changing the world".

You're probably a dumb teenager or a math major that hates people that actually use mathematics to perform work.

>ML is a meme
t. CS/Pure math memelord

I'd recommend understanding Calculus first, then move into Statistics.

If pic in this post doesn't make sense, then you'll want to review your calculus. MIT OCW and Khan academy are good for learning or reviewing calculus. The standard 'Calc for dummies" book should suffice if you want a fast review. Heard No BS Guide to Math & Physics (Calculus sections) might be a good review (haven't read it myself desu). But you can't go wrong with MIT OCW videos and supplementary practice problems.

Once you learn/review calculus look at:
projects.iq.harvard.edu/stat110/home


I don't mean Abstract Algebra, I mean basic algebra / trig /geometry as a pre-req to Calculus. Most of the difficultly with Calculus reduces to algebra / trig / geometry issues, not actual calc stuff. Calculus is trivial once you know how to do basic algebra / trig. It is purely mechanical. Analysis is where Calculus studies get interesting.

sauce on pic: neuralnetworksanddeeplearning.com/chap1.html

There's still a lot of work to be done in the Neuroimaging field you could apply these subjects to, especially things like fMRI data. These types of problems are already a hot topic but are only going to be increasingly popular over the next 5-10 years since we finally have the computing power to tackle them.