Back to index
OtherA full university course, roughly ten weeks of lectures with problem sets and a project, freely available as recorded lectures, slides and assignments·Free to follow via the public lecture videos and course materials, with a paid route if taken for credit through Stanford

Stanford CS234: Reinforcement Learning

4.3

A rigorous, university level treatment of reinforcement learning that sits comfortably alongside the other Stanford CS courses people already respect. It is demanding and unapologetically mathematical, but if you want to properly understand RL rather than just run an algorithm, this is one of the best free ways to do it, provided you bring the prerequisites.

What We Liked

  • Serious academic depth that builds reinforcement learning from its mathematical foundations rather than skipping to library calls
  • Taught by a genuine expert in the area, with a coherent progression from MDPs to modern deep RL
  • Lecture videos, slides and assignments are freely available, so the full course can be followed at no cost
  • The assignments are substantial, so if you actually do them you come out able to reason about and implement RL, not just describe it

What Could Be Better

  • Steep prerequisites in probability, linear algebra and Python mean it is a poor fit for beginners
  • Reinforcement learning is genuinely hard, and the course does not soften that, so the learning curve is real
  • The materials are less produced and less hand held than a polished commercial course, in the usual lecture recording style
  • Without enrolling for credit you have to supply your own discipline, since there is no grading or support around the free version

Detailed review

Stanford has quietly become the default source for serious, freely available computer science courses in this field, and CS234 is its dedicated reinforcement learning offering, sitting naturally next to the machine learning, deep learning, NLP and computer vision courses that people already point newcomers toward. Taught by Emma Brunskill, who works in the area, it takes reinforcement learning seriously as a subject in its own right and builds it up properly, starting from the foundations of Markov decision processes, value functions and the core algorithms, moving through policy gradients and exploration, and on into the modern deep reinforcement learning methods that connect the classical theory to today's practice. That coherent arc is the main reason to choose it, because reinforcement learning is an area where shallow tutorials leave you able to run someone else's code without any real grasp of why it works or when it will fail, and this course is built to close exactly that gap. The assignments are a big part of the value, since they push you to actually implement the ideas rather than nod along to them, and doing them is what turns watching the lectures into genuine understanding.

The honest caveats are the same ones that come with any real university course. The prerequisites are steep, and you will struggle badly without a comfortable grounding in probability, linear algebra and Python, so this is emphatically not a first course for someone new to the field, it is a course for someone with the machine learning basics already in place who now wants depth in RL specifically. Reinforcement learning is also just intrinsically difficult, arguably harder to build intuition for than supervised learning, and the course reflects that reality rather than smoothing it over, so the learning curve is real and you should expect to work. As with most academic material, the production is functional rather than glossy, these are recorded university lectures with slides, not a highly polished commercial course, and if you follow it freely rather than enrolling for credit you get no grading, no deadlines and no support, so the discipline has to come from you.

My take is that CS234 is one of the best ways to learn reinforcement learning properly and for free, but only once you are ready for it. If you already have the fundamentals and you want to move from being able to call an RL library to actually understanding the algorithms well enough to reason about and implement them, this is an excellent and rigorous route. If you are still early in your journey, get your machine learning and mathematics foundations solid first, then come back to this when you can give it the effort it demands and genuinely deserves.

[ final ]

The verdict.

The course to take when you are ready to understand reinforcement learning properly and are willing to do the maths and the assignments. Come in with solid probability, linear algebra and Python, treat the problem sets as the real work, and it will take you far further than any quick RL tutorial.