fast.ai: deep learning for people who already code

Most deep learning courses spend six weeks on theory before you train anything. fast.ai inverts that: you build a working model early and the explanation follows.

What you get

Part 1 is nine core lessons of roughly 90 minutes each, plus bonus material. Part 2 adds 16+ lessons going down into the foundations. The video lessons, the accompanying book, and the community forums are all free.

Coverage: computer vision, natural language processing, tabular analysis and collaborative filtering, plus random forests, regression and model deployment — worked through PyTorch, fastai and Hugging Face.

Prerequisites, honestly stated

“You know how to code (a year of experience is enough), preferably in Python,” and high-school mathematics. That is it. No specialised hardware is needed — the notebooks run on free cloud GPUs — and no large datasets of your own.

That last point matters if you are studying from Assam on a laptop. The barrier to this course is time, not equipment.

Who teaches it

Jeremy Howard, founder and lead developer of fastai, with 30 years in machine learning — formerly Kaggle’s top-ranked competitor, and its President and Chief Scientist.

No coding background yet? Start with Elements of AI.