Coursera: Mathematics for Machine Learning: PCA Review

Mathematics for Machine Learning: PCA is the third and final course in the new
Mathematics for Machine Learning Specialization offered by the Imperial College of London on Coursera. The course focuses on teaching linear algebra concepts related to principle components analysis or PCA: a common method to reduce the dimensionality of data. Although this is the final course in a specialization, it is mostly self-contained as it reiterates some the basics covered in the first course. Some of the quizzes and programming assignments are locked to audit students, so you'll need to pay for the certificate track to access all the content.


The first two weeks of Mathematics for Machine Learning: PCA focus on teaching foundational concepts such as mean, variance, covariance, and inner products while weeks 3 and 4 focus on projections and PCA itself. The decision to cover concepts like inner products and orthogonality is curious since those topics were already discussed in the first course of the specialization, but it does mean that this course works better as a standalone offering. The lecture quality and clarity are very good, but as with the other courses in the specialization track, it moves from basic concepts to advanced ones too quickly and presumes a level of mathematical maturity that may be beyond those whose only background is high school math. That said, the course should give you an understanding of how PCA works even if you don't follow all the math.


Mathematics for Machine Learning: PCA is a whirlwind intro to principle components analysis that tries to build up from the basics to PCA too quickly for its own good. The Mathematics for Machine Learning Specialization purports to be a learning track for beginners in the field of machine learning, yet each course jumps into fairly complex topics with screens full of notation, using jargon that hasn't been completely explained without spending enough time on the basics. The value of devoting an entire course to the mathematical underpinnings of PCA in a specialization geared toward machine learning beginners is dubious; for beginners understanding PCA at a conceptual level and being able to apply it with code is enough. A course covering important topics in statistics and probability would be far more valuable to the average user interested in math for machine learning, especially considering it would have been easy for the specialization to add a section on PCA to the first course.


I give Mathematics for Machine Learning: PCA 2.5 out of 5 stars: Okay.