While learning about the mathematical foundations of machine learning, I came across this lecture by Professor Gilbert Strang at MIT: Matrix Methods in Data Analysis, Signal Processing, and Machine Learning (18.065).

From the first video, I was enthralled.

In his way of teaching, Professor Strang takes this abstract and tedious class, to a new level. Concepts like basis, eigenvectors, null spaces, or SVD suddenly make a lot more sense. You don’t have to know how to calculate them, it’s “nice to have” but not a must. Most important is the ability to envision these abstract concepts and understand why and when to use them. At the end of the day, computer are better at computing than us and as Angus K. Rodgers once said:

Mathematics requires a small dose, not of genius, but of an imaginative freedom which, in a larger dose, would be insanity.

In addition, Gilbert Strang expresses a contagious enthusiasm for teaching the material and focuses on developing an intuition about it. Mathematical proofs are also taught, but from a procedural perspective that makes them much easier to absorb.

In summary, this course will equip you with an unparalleled ability to visualize linear algebra that is not easily obtained by taking a more traditional linear algebra course. Don’t think twice, take this course if you want to gain a deeper knowledge of machine learning. However, this course does require some basic knowledge of linear algebra. If you don’t have that or need a refresher on the subject, you might start by watching his other introductory course Linear Algebra (18.06).