Machine Learning Complete

The Machine Learning course from Coursera has ended. Well, it ended for me — assignments can still be submitted until August first.

I liked the course. I originally signed up because I wanted to use machine learning in a couple of my projects. Now I have a better understanding of what can be done with machine learning and what kind of results to expect.

Another impression that first arose from a university artificial intelligence course, and which repeated itself now — I’m surprised at how simple and understandable the ideas behind these algorithms are, backed by mathematical foundations. Take gradient descent, for example — that’s covered in the first year of university!

The course is primarily aimed at future users of various machine learning libraries, not at developers of custom algorithms. Hence the low entry requirements: no knowledge of probability theory, statistics, differential calculus and so on is required. I won’t claim that I remember all of that perfectly, but with more detailed optional proofs the course would have been even more interesting.

A few points on how the course was useful to me:

• Learned several machine learning algorithms and their application areas, got some new ideas.
• Figured out computation optimization through operation vectorization. A useful technique that would surely work great on modern GPUs.
• Learned about several libraries for machine learning and NLP. This wasn’t directly related to completing the course, but now I have an understanding of the principles behind how they work.

As for the downsides, I’d note the audio quality, which isn’t exactly bad, but is roughly on par with most Russian podcasts. You get used to it by the fourth week, but it’s annoying at first. And I would have liked deeper optional lectures.

And to make this post slightly more useful than just a personal opinion, here are some links:

• Another machine learning course, by the same author.
• Yet another machine learning course from the California Institute of Technology.
• A machine learning library for Python and an article analyzing the performance of different algorithms using this library. Actually, it’s a whole series of articles.
• An NLP library for Python.