Aug

7

The best data science course I have ever watched. In fact, probably the best data science course ever made is Prof. Yaser Abu-Mostafa's "Learning from Data" at Caltech. I am not exaggerating, and if you think I am, just read the comments section from the first video of the course The Learning Problem — its almost exclusively high praise.

This course is really old, as its from 2012. I watched it probably in 2017 or 2018. But its still very relevant today. Why its relevant today? Most courses focused on describing techniques that were popular then, but later became irrelevant. For example, GANs were replaced by Diffusion Models and core ML Architectures have shifted to Transformers.

Prof. Yaser's course is different because he covers Theory, Techniques and Concepts (most books/courses only describe ML algorithms/Techniques or how to use features in python libraries).

- Theory, refers to mathematical descriptions of ideas like "Is learning feasible" for your problem or dataset?, "Training vs Testing", "The theory of generalization" and the "Bias-Variance trade off".

- Techniques, refers to actual ML algorithms like Neural Networks, SVMs.

- Concepts, describes auxiliary things that are not really Machines Learning but useful to understand well. Like how to interpret/deal with Error & Noise, Sampling of data.

The full course is free.

I also recently came across a comment about a book he wrote to accompany the course which made me remember him: Learning From Data.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search