Should I gloss over the linear algebra chapter in the book "Deep Learning" by Ian Goodfellow? Currently I am reading "Deep Learning" by Ian Goodfellow, Yoshua Bengio, and Aaron Courville. I'm on Chapter 2 which is the Linear Algebra section where they go over the linear algebra that pertains to the book. I understand most of what is being taught but not at a deep level. And when I get to some of the latter parts of the chapter like "2.9 The Moore-Penrose Pseudoinverse" and specifically "2.12 Example: Principal Components Analysis", I don't really understand them that well at all.
Would it be okay if I go onto Chapter 3 and beyond before I understand these concepts comfortably, or will I be fine with having basic knowledge of them and the symbols they use?
 A: This is a question that often pops up when reading mathematical literature.
The initial chapters, of this book or any other math book, lay out tools that you will be using in later chapters, so strictly speaking, you will not understand the rest of the book without understanding these foundational chapters.
Realistically speaking, don't worry if you don't understand something. Continue reading until the topic actually appears and is applied. Then, and only then, re-read the earlier section, and try to make sense of it in the light of the later application. By then, you will have seen a lot of other material and may be able to understand it much better against this background.
In addition, it is often very good to look at other sources at this point, when you actually need to understand the application of something. Different authors have different ways of explaining stuff. Looking at things from different angles can be very helpful.
It has been said that good mathematical writing is the kind where you can mentally replace every formula by "foo" and still understand the gist. Read the formulas when you need to understand something in depth and detail.

Regarding the two specific topics you mention:


*

*The Moore-Penrose pseudoinverse is fundamental when you want to create an actual estimation algorithm. If you are mainly interested in applying algorithms someone else has developed and implemented, then you need to understand that algorithm, but much less so the gory details. I have never needed to understand the Moore-Penrose pseudoinverse. We only have very few threads here on it, too.

*PCA is much more useful to someone actually applying a tool. Conversely, someone building a tool will likely not use it very much. It's really good to understand this and related ways of reducing dimensionality or compressing information. If you come across a situation where PCA can be helpful in preprocessing, there will not be a big sign pointing this out, so you need to develop your own intuition and understand that this method exists. Happily enough, we have an astronomically upvoted mother-of-all-canonical-threads on PCA, along with an entire pca tag. Go through that thread, then re-read Goodfellow et al. on PCA. Enlightenment is almost sure to follow.
