It’s always the same question people ask: How much math do I need for Machine Learning? And the answer is: It’s really up to you, your projects and your goals. Getting started, you can use frameworks and libraries, that will cover most of the math for you, so that software engineering becomes a valuable skill, even without having a strong math background.
However, to be able to look at problems on a professional level, there is no way around having a proficient understanding of the math underlying Machine Learning and especially Deep Learning architectures.
Here’s what Andrew Ng has to say about it:
“While it’s hard to argue against knowing more math, I think the level of math needed to do machine learning effectively, or to get a PhD in machine learning, has decreased over the years. This is because machine learning has become more empirical (based on experiments) and less theoretical, especially with the rise of deep learning.”
With that being said, here’s a list of MOOCs that will give you just the right amount of math needed:
Improve your mathematics-based, analytic and outside-the-box thinking. This 10-week course focuses on the thinking processes required for mathematics.
Workload: 10 weeks, 10 hours/week
User rating: 4.8/5
This course is an introduction to Logic from a computational perspective. It shows how to encode information in the form of logical sentences, how to reason with information and it provides an overview of logic technology and its applications – in mathematics, science, engineering, business..
Workload: 10 weeks, 4-8 hours/week
User rating: 4.5/5
A mix of theory and practice at your own pace, starting with vectors and spaces, moving on with matrix transformations and alternate coordinate systems.
Where: Khan Academy
Workload: up to you
A classic. Professor Strang is awesome!
Where: MIT Open Courseware/YouTube
Workload: 34 Lectures
User ratings: 2.484.590 YouTube views, <11.000 upvotes (Lecture 1)
This textbook is aimed at math majors and graduate students, focusing on understanding the structure of linear operators on vector spaces. A variety of exercises in each chapter helps students understand and manipulate the objects of linear algebra.
User rating: 4.27/5
This course introduces visualization, probability, regression and other topics that will help you learn the basic methods of understanding data with statistics.
Workload: 8 weeks
This course provides an elementary introduction to probability and statistics with applications. Topics include: basic combinatorics, random variables, probability distributions, Bayesian inference, hypothesis testing, confidence intervals, and linear regression.
Where: MIT Open Courseware
Workload: up to you
This course covers data collection, analysis and inference, data classification, conditional probability, Bayesian modeling and basics of Linear Regression.
Workload: 5 weeks, 7 – 10 hours/week
This is an absolute classic and mandatory for Stanford students. Here’s what you get: Linear Methods, Classification, Basis Expansions and Regularization, Kernel Smoothing, Model Assessment and Selection, Model Inference and Averaging, Boosting, Neural Networks, SVMs, Random Forests, Unsupervised Learning, Ensemble Learning, High Dimensional Problems, …
Up until now, this has been the bare minimum you need. Now you can check out more difficult and specific math needed: Standford’s Machine Learning lectures (taught by Andrew Ng in 2008) are pretty math heavy. If you have managed to through all of the above you might be ready for this series.
Workload: 20 Lectures
User ratings: 1.702.826 views, <8.000 upvotes (Lecture 1)
Calculus (+ multivariate calculus)