Quantcast
[ 3 / biz / cgl / ck / diy / fa / g / ic / jp / lit / sci / tg / vr / vt ] [ index / top / reports / report a bug ] [ 4plebs / archived.moe / rbt ]

/vt/ is now archived.Become a Patron!

/sci/ - Science & Math


View post   

[ Toggle deleted replies ]
File: 126 KB, 1131x622, 1575300954085.jpg [View same] [iqdb] [saucenao] [google] [report]
11218013 No.11218013 [Reply] [Original] [archived.moe]

How much math should I take during my CS degree if I want to work with ML/AI after graduating.

>freshman who's taken calc and physics 1, planning courses for next semester + the rest of my degree.

>> No.11218039

Linear algebra and calc 1 should be sufficient for an intro course. Statistics will help for theoretical courses

>> No.11218041

>>11218013
brainlet level CS, do TCS so you can take as most math as possible, a few examples being Combinatorics, Abstract algebra, group theory, Commutative ring theory, Representation theory, Commutative algebra, algebraic geometry, Number theory, lambda calculus, Computability theory, Graph theory, calculus of finite differences, discrete calculus, discrete analysis, Set Theory, Probability, topology, Game theory, decision theory, utility theory, social choice theory, Discretization, Operations research, Discrete analogues of continuous mathematics, Hybrid discrete and continuous mathematics, complex analysis, real analysis, Automata Theory, Differential Equations, Stochastic Processes, and Discrete Geometry.

>> No.11218046

Linear algebra, differential calculus and optimization is all you need to work in the industry.

If you want to go deeper you should take probabilities and functional analysis to understand the proofs on universal approximation of deep networks, the effectiveness of the lasso estimator for sparse regression (ie when you have more features than observations) and generalization properties of algorithms in ML.

Good luck anon

>> No.11218048

>>11218013
Start with vectorial calculus and basic linear algebra, at least you can learn perceptrons with that alone

>> No.11218054

>>11218041
>combinatorics
gay
>abstract algebra
gay
>group theory
gay outside of physics
>commutative ring theory
gay
>representation theory
gay outside of physics
>commutative algebra
gay
>algebraic geometry
gay
>number theory
you won't contribute anything to this field
>lambda calculus
brainlet
>computability theory
not math
>graph theory
trivial
>calculus of finite differences
an abomination
>discrete calculus
see above
>discrete analysis
see above
>set theory
not math
>probability
you won't understand it
>topology
see above
>game theory
not math
>decision theory
see above
>utility theory
see above
>social choice theory
see above
>discretization
gay
>operations research
not math
>discrete analogues of continuous maths
proof-read your posts
>hybrid fake and gay
see above
>complex analysis
lol
>real analysis
omega lol
>automata theory
not math
>differential equations
lol
>stochastic processes
lol
>discrete geometry
extremely gay

>> No.11218066

>>11218054
1) there are different topics than just these in TCS
2) yours is the most “I’m an analysisfag and no other math is actually math” post out there
t. analysisfag

>> No.11218067

>>11218013
why is CS so ceaselessly gay and self conscious? is there another analogue to this kind of pure cope other than analytical philosophy?

>> No.11218068

>>11218054
OK tenure boomer

>> No.11218070

>>11218066
those are also usually not math, gay or lolworthy distortions of those topics

>> No.11218085

>>11218054
>complex analysis
>lol
wrrong, it's "gay outside of physics"
do you even?

>> No.11218090

just become an "applied ML" person and use the easiest copy and paste shit. Get lots of experience. Leave the math to nerdfaggots and BS on your resume then try to work up a corporate ladder.

Math is for fags

>> No.11218092

>>11218070
Nah. CS research isn’t stuck in the 70s

>> No.11218098

For a basis:

Linear Algebra (really just affine functions)
Calculus
Optimization (specifically gradient methods)
Probability/statistics
Information theory
Numerical methods

For all of the above, you don't actually need to learn the entire field. You just need the subsets that apply to machine learning. The first two sections of the Goodfellow book give a good idea of the areas to focus on (don't bother with the third section, it's dated at this point).

www.deeplearningbook.org/

For ML/AI:

Linear models (regressions, trees)
Bayesian methods
Unsupervised clustering
Dimensionality reduction (PCA, tSNE, UMAP)
Deep models (MLP, CNN, RNN/LSTM)
Modern architectures (resnets, transformers, etc)
Regularization
Generative models (images, text)
Outlier detection

I would go with Andrew Ng's course for traditional ML and fast.ai for deep learning.

This will give you a decent base, but the place you need to get to if you want to get hired (especially with just a BS) is to be able to read, understand and implement papers fairly quickly. Also in terms of learning, actually coding stuff up is many times more valuable than reading about stuff.

>t. machine learning research scientist

>> No.11218113

Start working on ML and AI. If you need math learn the math as you go.

>I am going to learn X so when I learn Y it makes sense
In this case you won't be sure what from X you need, if it applies, and you might learn a weird abstraction of it that is less applicable to Y

>I'm going to learn Y and specifically learn X as needed
This is more efficient. You will learn X as needed and in a directed way towards exactly what you need and you will know the context it is used in.

what you are doing is like this
>I am going to go to Asia in 3 years to live in a foreign country
>I am therefore learning "Asian Languages" for the next three years so I can know some words for when I go there

SEE THE PROBLEM?

You can just start learning AI and ML right now and then GO BACK to the math you need as needed. Which is infinitely more efficient.

Aka
>Go to Asia right now and determine country you want to stay in
>learn that language while immersed

seems a lot more intelligent right?

So

A) Begin Learning AI and ML right now.
B) go over the necessary math as you need it.

The one downside is you will see things you don't understand at all and if you are a weak bitch it might be intimidating.

>> No.11218118

>>11218098
>I would go with Andrew Ng's course for traditional ML and fast.ai for deep learning.

this is exactly where you should go right now. Not through 10 different math books you might need something from.

>> No.11218348

>>11218113
>You can just start learning AI and ML right now and then GO BACK to the math you need as needed. Which is infinitely more efficient.

>> No.11218414

>>11218054
>combinatorics
>gay
>abstract algebra
>gay
>group theory
>gay outside of physics
>commutative ring theory
>gay
>representation theory
>gay outside of physics
>commutative algebra
>gay
>algebraic geometry
>gay
These aren't arguments
>number theory
>you won't contribute anything to this field
https://www.sciencedirect.com/science/article/pii/0898122182900268
Normal Computer scientists contribute to this field let alone Theoretical computer scientists lol
>lambda calculus
>brainlet
Ok than solve this problem.
Without using recursion or a fixed-point combinator, define a function as even which, when applied to a Church numeral, returns the Church encoding of true or false, depending on whether the numeral represents an even number or an odd number.
>computability theory
>not math
Mathematical logic is math bro
https://en.wikipedia.org/wiki/Computability_theory
>graph theory
>trivial
If graph theory is so trivial than you can solve this problem Show that any graph where the degree of every vertex is even has an Eulerian cycle. Show that if there are exactly two vertices a and b of odd degree, there is an Eulerian path from a
to b. Show that if there are more than two vertices of odd degree, it is impossible to construct an Eulerian path.
>calculus of finite differences
>an abomination
>discrete calculus
>see above
>discrete analysis
>see above
These aren't arguments

>> No.11218416

>>11218414
Continuation from previous post
>set theory
>not math
lol what set theory is mathematical logic, are you really going to say it isn't math. https://en.wikipedia.org/wiki/Set_theory
>probability
>you won't understand it
>topology
>see above
Maybe a brainlet undergrad CS students, but if you go for TCS in grad school or a PHD in TCS you have to learn probability and topology
>game theory
>not math
Oh come on studying things mathematically is math. You're not going to tell me physicists do no math now are you? https://en.wikipedia.org/wiki/Game_theory
>decision theory
>see above
>utility theory
>see above
>social choice theory
>see above
lol I'll give you these, these aren't math
>discretization
>gay
Not an argument
>operations research
>not math
Kinda iffy on this one, it seems to be a subset of applied math but its kinda vague so I'll give you this one.

>> No.11218431

>>11218416
>>11218414
Third Continuation
>complex analysis
>lol
>real analysis
>omega lol
Not arguments
>automata theory
>not math
lol sure, heres the mathematical foundations of automata theory if you think it isn't a math.
https://www.irif.fr/~jep/PDF/MPRI/MPRI.pdf
>differential equations
>lol
>stochastic processes
>lol
>discrete geometry
>extremely gay
Not arguments

>> No.11218433

>>11218013
don't take any math, it's all a bunch of math. focus on the coding.

>> No.11218436

>>11218414
>>11218416
>>11218431
Final Continuation
>discrete analogues of continuous maths
>proof-read your posts
>hybrid fake and gay
>see above
Ok now you're being a supreme brainlet.
Continuous concepts in math have discrete versions, an example being discrete modelings with is a discrete analogue for continuous modeling. Hybrid discrete continuous models are just what they sound like.
Just because you don't know what some things are doesn't mean they're fake or that I made a mistake.

>> No.11218442

>>11218436
you have autism, not the useful kind either.

>>
Name (leave empty)
Comment (leave empty)
Name
E-mail
Subject
Comment
Password [?]Password used for file deletion.
Captcha
Action