What do I need to know before beginning Deep Learning beside stats and linear algebra?

Skip to content
# What do I need to know before beginning Deep Learning beside stats and linear algebra?

###

What do I need to know before beginning Deep Learning beside stats and linear algebra?

Can someone explain why linear algebra is used in machine learning? I thought it was mainly used for geometry.

just basic matrix math enough

information theory helps here and there, but not really that essential

pca and eigen decomposition is about the most complex thing you can practically find beyond matrix operation

>mainly used for geometry

Oh god, no, far from it.

It's mainly used EVERYWHERE, and I'm not exaggerating. Linear spaces are such versatile algebras that you stumble everywhere into something that you can pack up in the abstract suit of a vector space and use all the tools given to you by linear algebra. The first thing that comes to my mind where you will find it most useful is actually differential equations, not geometry. But you will find a lot of it in mechanics, circuit analysis, signal processing, graph theory... the list does on. You can basically pick a random branch and you'll find out they have some fundamental uses in it

>I thought it was mainly used for geometry.

Lol no. Linear algebra is everywhere, from statistics to signal processing.

machine learning literally just taking the variables and the result and solving for the equation.

Negro Linear alegebra is mainly used for CFD and FEM.

The rest of the math

What you've posted isn't actually that difficult. It looks forbidding because there's a lot of symbols, but it's just fairly basic stats plus a lot of sums and logarithms.

I don't know about that I see a sideways trident symbol some italic h, I think that funny o is phi or something and a lambda calculus looking thing... All in all it says it's multivariable calculus otherwise known as calc 3

I don't know what calc 3 is but I don't see any calculus there. It's just a bunch of sums and logarithms.

The sum notation itself is calculus

Nah, it's just basic math.

Well the source says its multivariable calculus or whatever.

https://www.datasciencecentral.com/the-mathematics-of-machine-learning/

I think the author meant not that this particular screenshot shows multivariable calculus, but that you need multivariable calculus in general to understand machine learning. Which is true, though many people get by without understanding. Overall I agree with him, though I would change the order a bit to make stats #1 and I'd throw out some things from #5.

If you are interested about machine learning in practice, you don't need linear algebra. You need stats and python.

unfortunately it's a ""math""" master degree where almost everyone does a phd after

>You need stats and python.

retard take.

he needs to actually know what he's doing, or he'll just be copy pasting examples off the keras website and randomly tweaking parameters by trial and error and wasting hours and hours of resources and training time

>t. took deep learning class in college after not paying attention in math classes and that's exactly what ended up happening

Nobody knows what they are doing because the practice is way ahead of the theory. You don't know what the software is really learning and you don't know why it works so well. You only know it works and you publish your results to show other people you did it first: most research papers are just "this thing applied to this problem and here's our results". You can learn all the math you want but it won't teach you how to shape the network and how to tune the hyperparameters. It's just a matter of practice, trial and error. All the cool stuff comes from very few people and the usual big corporations. Even when they discover a significant breakthrough they barely understand the implications of what they did:

>Somewhat surprisingly, it was found that similarity of word representations goes beyond simple syntactic regularities. (word2vec paper)

>Somewhat surprisingly...

Let's cut to the chase

How much math would I need to make dallemini make porn

Not much, if at all, just need to know how to program

That you will be a corporate slave for life if you specialize in this.

Not OP, but I already am a corporate slave. Would rather be a corporate slave who makes more money though

Nothing, really. Deep learning isn't very technically challenging, the skill is mostly an art where you intuitively learn to tweak hyperparameters

Bayesian optimization

Seriously, anything around "deep learning" turns around that.

Not OP but tell me more. How is the Bayesian approach to stats useful for neural networks? How does deep learning turn into it? I've never seen MCMC and related techniques used there.

Start with shallow learning and work your way down

Why is it so much harder to sit down and self teach math than programming...

Must be because learning math requires 0 creativity and it BORING

>math requires 0 creativity

You're doing it wrong.

import deep-learning