How to develop mat. apparatus to the minimum level for machine learning and data science?

Question

css

In order to engage in serious and analyze machine learning, in general, you need to have a good mat apparatus.
Ruler, statistics - how to develop these things to the minimum required level? so as not to learn unnecessary abstract things like manifolds, but exactly what you need?

Answer 1

Use Radial Gradients Luke.
Well, or at worst, add pseudo-elements.

Answer 2

looks like chewing gum :D

Answer 3

Horror, the button from 2010 came along the way.
If you try to do this with gradients, pseudo-elements and stuff like that, and a big-eyed designer will accept it, then you will go crazy, I guarantee you that. So it's easier to make it a picture (without a number, of course).

Answer 4

oh yes, you don’t need any apparatus there, you need to understand how to multiply / divide matrices (conditionally), well, how to calculate the probability, there you can master all the necessary concepts and tools in just an hour, then you solve applied problems, look how others solved them - every day of the year 2-3 so you decide and become more or less a specialist.

Answer 5

It is for the minimum required level that there is not so much needed. Pay attention to the book, which, in my opinion, is both a good launch pad for DS in general, and may be relevant, in particular, in the context of your question: Principles of Data Science

Answer 6

Learn to read formulas.
These tables of mathematical signs are on Wikipedia.
Then - you will understand: what they write about in publications on ML.
And after - you will design the algorithm.
And only at the very end - to code.

Answer 7

All data science is based on terver and statistics. In a particularly perverted form.
Information theory is needed for at least some understanding of the work of neural networks and other deep learning. It is necessary to dig here starting from information bottleneck theory by Naftali Tishbi et al. But this is not accurate :-)
Linear algebra and matan are used exclusively utilitarian, at the level of matrix/tensor multiplication and derivative/gradient calculation. And also for understanding the principles of parametric optimization ("nonlinear programming") in general.
And, of course, the basics of algorithms and data structures, and their asymptotic analysis.
On all educational platforms like edx, coursera, udacity, there are now "specializations" (micromasters, nanodegree) on the topic of data science, which just include everything you need, and precisely in the minimum required volume. Sometimes even much less, but they set the direction.