Linear Algebra For Machine Learning
Over the summer I took a UC San Diego Extension course entitled “Linear Algebra for Machine Learning”. The purpose of the course was to apply common linear algebra topics to machine learn. I recently completed a Linear Algebra course of Pasadena City College (PCC) in the fall. I thought this course would be a good opportunity to practice some of the concepts I had previously learned to highly sought after skill set.
Topics covered:
- Tensor operations and their role in data representation and early neural network models such as MLP
- Expressing systems of linear equations in matrix format and obtain the solution set, equivalent to calculating weights of neural network
- Calculate basis for a vector space, change of bases, used for text and image linear transformations: projection, mirroring
- Solving linear regression with matrix inversion of non-square matrices
- Continuous neural network optimization using gradient descent
- Singular value decomposition (with applications in image compression, reconstruction)
- Dimensionality reduction & principal component analysis
- Tensor flow 2 regression models
Class format:
Weekly lecture videos were posted to canvas. The lectures each week were long. Somewhere between 2 to 3 hours. Along with lecture videos there were weekly discussion post and quizzes. Some times there would be additional optional statistical analysis projects. The projects used Octave, a scientific programming language. I had not heard of this language before this class. Its fairly easy to pick up if you have experience with Python. This class only uses a few commands. Overall, the projects are a good way to practice some of the topics taught in lecture.
Assignments:
The main assignment outside of the quizzes are the discussion post. You are required to post one and respond to two other students post. Posts could range from casual spoken language to more formal mathematical proofs. There was really range of students in the class with varying levels of math experience. I found that I was on the more experience side given that I had already taken a linear algebra course. I liked to respond the discussion questions with a lot of formal mathematical notations. I thought it helped me expression my thought more clearly and thoroughly.
Exams:
There are weekly quizzes. I found these quizzes a lot easier than exams I took at PCC. They really only require one to have a basic understanding of topics and did not require complex calculations. You did not have to do proofs on quizzes.
Overall:
Overall, I felt like the course was just getting into the machine learning. At the same time I felt the course rushed through the linear algebra concepts. I was glad that I had already taken a linear algebra course. I imagine that all the students who had little math exposure really struggled. I would only recommend this class if you have previous college level math experience. I think the ideal target audience for this course are graduate level STEM student who are maybe interested in machine learning for their research, and want an introduction to the fundamental concepts. If you are a professional who has been in the work force for a number of years and the last math class you took was high school algebra your are going to be overwhelmed.
