Steve Brunton is a mechanical engineering professor at the University of Washington, and his YouTube channel (Eigensteve) has become the default reference for a corner of machine learning that the big MOOCs barely touch. The core of it follows his textbook with Nathan Kutz, Data-Driven Science and Engineering, and moves from the singular value decomposition through Fourier and wavelet methods, sparse regression, dynamic mode decomposition and SINDy, into control theory and reinforcement learning. More recently he has added a physics informed machine learning series that walks through the stages of building a model (choosing what to model, curating data, designing an architecture, crafting a loss function) before covering PINNs, Fourier neural operators, DeepONets and Hamiltonian networks. What makes it work is the teaching.
He writes on a lightboard facing the camera, talks at a steady pace and keeps returning to why a method exists rather than just how to run it. I have watched his SVD videos several times and they are still the clearest explanation of the topic I know, better than most university lectures I sat through. I also like that the physics informed series is honest about where PINNs fall short, which is rare in a field full of breathless papers. The limits are worth being clear about.
This is applied maths for scientists and engineers, not an AI career course. You will not learn anything about prompting, RAG or deploying LLM apps here, and there is no certificate to put on LinkedIn. The playlists are not sequenced as a single course either, so you need some self-discipline, ideally with the book open beside you. Some of the older material assumes MATLAB, although the Python port of the book code helps.
For the right person, though, nothing paid comes close. If your work involves fluids, structures, robotics, energy systems or any physical process, this is the free series I would recommend first, and the book is one of the few ML textbooks I would actually pay for.