From ML class, what kind of math is this?

>>8650083
It's mainly probability theory.

t. phd student

>>8650083
interuniversal teichmuller theory

>>8650086
Also just wanted to point out that machine learning always looks like this. If you want to just learn some shallow theory behind how to implement a learning algorithm, do not take a machine learning class. Maybe an undergrad one isn't as bad but at the graduate level it's pretty hardcore.

>>8650103
Makes sense. I don't know any probability theory. Where should I start & what pre-reqs should I have before I start with probability.

>>8650110
There's a mix of probability theory, multivariate calculus and linear algebra (at least up to SVD). You need all of that before you can really start learning ML.

>>8650083

Computer science.

Isn't it
[eqn]
\begin{align}
\frac{1}{N}\sum\limits_{i=1}^{N}\log\frac{f(\mathbf{x}_i)}{p(\mathbf{x}_i)} &= \frac{1}{N}\sum\limits_{i=1}^{N}\log\frac{ \frac{1}{N}\sum\limits_{j=1}^{N}\mathbf{1}(\mathbf{x}_i = \mathbf{x}_j )}{p(\mathbf{x}_i)} \\
&= \frac{1}{N}\sum\limits_{i=1}^{N}\log\frac{\frac{1}{N}}{p(\mathbf{x}_i)} \\
&= \frac{1}{N}\sum\limits_{i=1}^{N}\left(\log\frac{1}{N} - \log p(\mathbf{x}_i)\right) \\
&= \frac{1}{N}\sum\limits_{i=1}^{N}\log\frac{1}{N} - \frac{1}{N}\sum\limits_{i=1}^{N}\log p(\mathbf{x}_i) \\
&= \log\frac{1}{N} - \frac{1}{N}\sum\limits_{i=1}^{N}\log p(\mathbf{x}_i)
\end{align}
[/eqn]

>>8650155
I don't see any computer science in that image.

>>8650157
You can never really tell with random ML excerpts because people introduce their own notation in every other paper.

>>8650159
harr harr, then why is it offered as a graduate CS course?

>>8650159

Well, that's pretty much all I did in my computer science courses.

It's applying mathematics for a niche purpose - computer science.

>>8650159

Let me give you an analogy. Would you call physics math just because physics uses math as a tool?

Then why would you call computer science math just because we use math as a tool?

Non-sequitur.

>>8650171
Right idea, wrong application. Statisticians use computers and computer programming as tools. That does not mean that statistics is computer science.

>>8650182
Statistics isn't quite ML. Which you would know if you ever took a ML class. ML uses probability theory as one of the main tools however.

File: Screen Shot 2017-02-03 at 9.22.12 PM.png (65KB, 839x512px) Image search: [Google]

65KB, 839x512px

>>8650131
Ok thanks.

>>8650159
This is a grad level CS course. Haven't seen any applications yet.

>>8650195
>Statistics isn't quite ML
Do elucidate the differences for us.

"""Machine learning""" is a buzzphrase moniker for prob/stats used to secure research funding, get dumb undergrads to take your class, and spam in your startup's bio to ensure that you're bought-out.

>>8650216
>>8650083
If that is what ML really looks like, then that looks boring. Who is really interested in that shit? There are more satisfying areas of math than statis/probability shit.

>>8650229
In statistics you study distributions to understand the structure and properties of them.

ML is about learning an unknown function given a sample from an unknown distribution. As you can imagine there is some overlap which is evident in the way we model the problems.

Some statistical techniques work well in ML, such as linear regression. Others are purely ML such as neural networks.

But what ML entails is exactly what I described, that you try to learn an unknown function of observed data.

>>8650245
This is the theory behind ML. If you want to gain wizard status then you need to understand the theory child.

Of course you can always use a library and try experimenting on real datasets.

>>8650253
yeah i guess ill take the effort to learn the theory. id rather do that than just use libraries

>>8650229
Are you just going to slink away now that your low effort trolling has exposed your ignorance?

Filthy undergrad.

>>8650264
Using libraries is a good thing and a lucky part of being programmers as well. It's nice to be able to actually see results and get some tangible motivation to dive deeper into the theorems.

>>8650267
>>8650250
>In statistics you study distributions to understand the structure and properties of them.
>I'm the ignorant one

>>8650270
quite true my man. i completely agree. you're getting me motivated again to learn ML.

>>8650272
Pray tell undergrad-kun, what do you do in "real" non-interpretive statistics?

>>8650272
Are you too pajeet to understand the validity of my generalization? It's okay, I understand.

>>8650280
Quantify uncertainty.

>>8650157
Very discrete
Much wow

>>8650302
You're thinking more of probability than statistics, which in either way is not the same as ML which I pointed out the difference for you.

It's funny how much bullshit people throw at ML who know nothing about it except whispers made by nervous outdated professors worried about their own funding.

>>8650302
Also fuck you. You shit on my perfectly reasonable definition of the problem of statistics and then you respond with the ***most*** undergrad-ass definition possible.

>>8650309
Statistics and probability are two sides of the same coin, "undergrad-kun".

Take a look at some current research
https://arxiv.org/pdf/1701.07875v1.pdf

Clearly by your own admission this had to do with statistics since it's concerned with the properties of a measure of distance between distributions. To prove these properties, however, the authors must delve deep into measure theory and Borel sets, topics typically categorized under fundamental probability theory.

>>8650328
>In statistics you study distributions to understand the structure and properties of them.

Ask yourself this child, does anything you say refute anything I said?

>>8650339
Yes "child" it refutes your insistence that statistics and probability are somehow different. And since your retard definition doesn't mention probability, you clearly don't understand either.

>>8650328
Oh and just to play the "lets google a random paper" game here's a paper on CNNs:

https://people.eecs.berkeley.edu/~jonlong/long_shelhamer_fcn.pdf

Surely this is just statistics right? Not worthy of it's own field of course.

>>8650346
How are distributions defined? As PDFs. Do you really need someone to spell out everything for you?

You asked me to tell you how ML is different than statistics, which I did.

After being BTFO you are now ass devastated and going off on bullshit tangents that have nothing to do with the topic at hand.

>>8650346
Another thing child, if statistics and probability weren't different, why are they two different fields?

>this will be entertaining

>>8650354
>>8650349
>How are distributions defined? As PDFs
rofl

As I thought you're just a neural networks fanboy that thinks shitting out yet another network architecture to chew on yet another labeled dataset to increase a benchmark by some small margin constitutes a major contribution and makes you a big shot.

Yeah, that's it's own field, it's called "garbage".

>>8650365
mmm the salty tears of an undergrad. Delicious.

>>8650365
Hopefully one day you will understand that being open minded to other fields is a sign of intelligence. I understand that undergrad is an isolating experience that's very hard and scary but learning to be receptive will be a great asset in life.