AI and ML learning path

The courses I would take, in order, to go from curious to doing machine learning work, by pctablet505, an ML engineer working on on-device AI (models that run on the phone rather than on a server). Course details checked on 21 October 2026; there is no total time for the route, since several course pages state none.

Start with CS50's Introduction to AI →

1. A first look

Start here to see the whole field before you choose a part of it. After it you can explain search, learning and neural networks, and train a small model. Skip it if you have trained one before.

• CS50's Introduction to Artificial Intelligence with Python (Harvard): free, seven weeks of material. It expects CS50x, Harvard's introduction to computer science, or a year of Python.
• Machine Learning Crash Course: a fast-paced, practical introduction with videos and exercises. Its page states no cost or length.

2. The foundations

Everything later builds on these algorithms and the mathematics behind them. After it you can write the core algorithms yourself, without a library, and follow how a model is trained. Skip it if you have passed a university machine learning course.

• CS188, Introduction to AI (Berkeley): one semester of graded projects (the site is the Fall 2018 run). It expects data structures and algorithms; effort gets you through.
• MIT 6.036, Introduction to Machine Learning: 13 weekly units online, with more mathematics and each algorithm written by hand.
• Machine Learning Specialization (Andrew Ng, Coursera): three beginner courses, about two months at 10 hours a week, explained with a lot of patience. It is paid; its page says a free audit, taking it without paying, gives no certificate.

3. Go deeper, in one direction

Pick a direction by interest; after it you can work in it on your own. Skip the one you will not use.

• MIT MicroMasters in Statistics and Data Science, a set of MIT online courses: four on a track (focus area) you pick, always with Probability and Machine Learning, then a capstone exam (a final exam on the program, watched online). It lists pathways into graduate programs, but no cost. Statistics is hardest.
• DeepLearning.AI: start with the Deep Learning Specialization, then NLP, computer vision or generative AI. Cost and length differ by course.

4. Side roads worth taking

Each of these goes deep in one area once the basics hold.

• CS231n, Deep Learning for Computer Vision (Stanford): one quarter; the 2017 lectures are on YouTube.
• CS285, Deep Reinforcement Learning (Berkeley): one semester, with the weekly schedule on the site.
• MIT 18.06, Linear Algebra (Gilbert Strang): free on MIT OpenCourseWare, as taught in spring 2010. AI without mathematics is like a Lamborghini without an engine: it's cool, but it doesn't work.

After the courses

After about a year of courses you will feel ready. Then pick one problem, your own or one everyone has, and work on it for a year or two. You will find that training the model is less than a tenth of the work; the rest is deciding where machine learning fits, and turning the problem into one it can solve.

Jobs in AI and ML (my view, not a survey)

Research roles usually ask for a PhD; a master's followed by a PhD, or a direct PhD, is the usual route. Engineering roles (AI developer, ML engineer, data scientist) do not need one, and are open at startups and large companies alike: keep applying, and keep meeting people.

---

_{Built by pctablet505.}