>>10846994That's because ML and other what you called advanced statistics brings a lot of baggage from different fields. Like applied linear algebra and such. On a surface level, I think Bishop and Hastie are good first read. Then you should stick on an algorithm you want to learn in deep and study what it use.
I think it would be unwise to study statistics from bottom-up if your goal is to understand ML algorithms. That's because lots of low-key statistics wouldn't have a first definite use on what you understand as ML algorithm, like sufficiency or ancillary statistics. Heck, I don't even know if these matters on ML.
I think there's some tree to learning statistics. I would call them math-heavy branch, as the usual core until measure theory (and maybe beyond), and some linear algebra; the applied math branch, covering the numerical tools and algorithms of mathematics and computer science; and the statistics part.
By statistics, it would be wise to start with Inference. For this, I would recommend, in order:
*Probability and statistics by Morris H. DeGroot: as usual with inference books, cover some basic probability in the start. Good intro to various topics that will be explored in the following books.
*Statistical Inference by Casella & Berger: expands concepts of the other book.
*Introduction to the Theory of Statistics by Alexander M. Mood, Duane C. Boes, and Franklin A. Graybill: I think it's the base ready of Inference courses. It covers well with rigor.
These are all basic books developing the base of statistics, what is called Inference. From now on, after learning these subjects and the what is now called classical or frequentist approach, you would be suitable to study some other stuff.
