Need advice getting into machine learning (or Big Data, at least)...

>>8828790
Learn probability and statistics

>>8828790
These pictures are so dumb. DNNs actually look like this:

[eqn]y = f_n\left(W_n \cdot\ \cdots f_3\left(W_3\cdot f_2\left(W_2\cdot f_1\left(W_1\cdot x + b_1\right)+b_2\right)+b_3\right) \cdots +b_n\right)[/eqn]

>>8828826
It's like conical topology for black holes, it is for the layman to understand.

>>8828826
This is the reason I hate ML researchers. Most of them just care about showing everyone else how good they are at linear algebra.

Their presentations are full of this garbage, as if the audience is actually going to read a slide covered in equations.

>>8828894
Have you considered that those presentations are not intended for you.

>>8828790
Write a C program that count number of occurence of words as it read from input stream.
If you can write this they you probably have good foundation.

>>8828905
These talks I went to weren't at a fucking ML conference. They were for a general CS grad audience.

The point is that ML people don't know how to give any other type of presentation.

>>8828913
Can you give an input/output example of this problem?

>>8828826
That is a super clear explanation my man.
Every freshman will understand what these undefined functions mean

>>8828937
f_i are pointwise non-linearities e.g. relu, sigmoid, tanh
W_i are matrices
b_i are vectors

>>8828916
>ML people don't know how to give any other type of presentation
They don't know how to make a decent presentation in general. At my uni, we had two guys in charge of CS and they made really lousy presentations because they never were taught different. No one can tell them they suck at it because they're the top dogs and telling them anything like that is a way to get on their bad sides and never graduate (or lose your job).

>>8828894
>Most of them just care about showing everyone else how good they are at linear algebra.

If you don't understand linear algebra then you really shouldn't expect to understand neural networks. It'd be like explaining Calculus to a person that can hardly count.

>>8828894
>ML Researchers show pictures
>"WAAAH STOP TRYING TO PRETEND YOUR SILLY MATRIX MULTIPLIES ARE LIKE THE BRAIN"
>ML researchers write equations
>"WAAAHH STOP TRYING TO CONVINCE EVERYONE YOU ARE GOOD AT LINEAR ALGEBRA"

stfu faggot

>>8828790
Do you want to get into neural networks? That's fairly easy as there are multiple libraries available and fully encapsulated online tutorials. There's not much math to it either beyond basic linear algebra. The downside is that theres nothing really there that hasnt been there for almost 30 years. The resurgence in neural networks is completely technology based. There really isn't anything there.

I'd recommend going through the classic classifier progression. Fischer -> PCA/ICA -> SVMs. Then look into generative models + graphs. Finally, tackle more recent topics that are math heavy like manifold learning. Neural networks are their own thing that you can cover at any time (knowing graphs helps but isnt necessary unless you really want to get into RNNs)

Or you can just read Bishop. It's been the machine learning bible for as long as I can remember. And its free.

Why are people still obsessed with neural networks? support vector machine are superior in every single way (except in the 'OMG its like a brain you are making in a computer!' part).

>>8829477
>still using SVMs
It's not 2011 anymore gramps.

>>8829477
Not every problem is classification.

>>8829345
Well, deep learning in general. I've read a good portion of Bishop and Mitchell. I kinda just want to put it into practice, preferably using one of those technologies that are hot these days like Spark (since I know Scala and want to improve my chops), so I can transition into that kind of a job.