7
$\begingroup$

I'm looking for a Machine Learning course that would give the maths behind algorithms rather than simply teach how to apply them. I've looked at Udacity Into to Machine Learning and Andrew Ng's course on Coursera, and they both seem too applied to me. Any recommendations of textbooks would also be much appreciated.

$\endgroup$
5
$\begingroup$

To add on to @Digio, I would recommend Abu-Mostafa's Learning From Data, which contains enough statistical learning mathematics to get you excited and wanting more.

$\endgroup$
5
$\begingroup$

Note, Andrew Ng has a more mathematical course in Stanford Online not Coursera.

Recommendations would always be subjective, for me, I personally like

The Elements of Statistical Learning

Convex Optimization

Both books are classical books in machine learning community and freely available.

Related question can be found here.

Machine learning cookbook / reference card / cheatsheet?

$\endgroup$
4
$\begingroup$

Try to dig deeper in a specific topic. Ngs course only scratches the surface, but other more specific courses are more theoretical/mathematical.

Bayesian networks/ Markov networks:

Probability graphical models is a theoretical advanced coursera course on Bayesian networks/Markov networks. The book is even more theoretical. It is full of proofs.

Neural Networks:

Neural Networks for Machine Learning is also a rather theoretical course as it is really profound. Nonetheless it is not as mathematical as the PGM course I mentioned above.


However if you want to understand the theory of Machine Learning itself and not the algorithms you can go for a textbook. In this case go for what @digio proposed.

$\endgroup$
3
$\begingroup$

This (archived) edX machine learning course from Columbia explains a lot of underlying math. For example they show regularized linear regression and probabilistic matrix factorization from Bayesian (Maximum A Posteriori) perspective.

Understanding Machine Learning is a (freely available) textbook that takes computational learning theory approach, and contains derivations and calculations/estimation of VC dimension of classifiers.

$\endgroup$
  • $\begingroup$ is all of the math you mentioned still relevant in 2018/2019 or how has it changed? what type of math should i be learning as a base for machine learning right now? ive done some searching and the newest material regarding this that i can find dates back to 2017, and i know that ML has been growing in the meantime $\endgroup$ – BugWhisperer Nov 22 '18 at 5:00
  • $\begingroup$ @user2230470 I'd say that math for most machine learning didn't change in last 10 or 20 years. I mean math for basics, more specific models might need understanding some different things (for example there are lots of hard math in Generative Models, like Wasserstein distance in WGANs or Variational Inference in VAE) $\endgroup$ – Jakub Bartczuk Nov 27 '18 at 0:06
  • $\begingroup$ hasnt the popularity of activation functions changed from sigmoid to relu tho? ive started learning linear algebra. multivariable calculus is another one that ive heard. it would just be nice to understand the different types of mathematics that are generally used in the field $\endgroup$ – BugWhisperer Nov 27 '18 at 9:27
  • $\begingroup$ I don't get what activation functions have to do with that. They're defined using arguably high-school math. $\endgroup$ – Jakub Bartczuk Nov 27 '18 at 9:32
  • $\begingroup$ don't the Sigmoid and ReLU functions involve different types of mathematics? $\endgroup$ – BugWhisperer Nov 27 '18 at 21:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.