Stanford Graduate Certificate in Artificial Intelligence

Learning plan

About 680 hours of focused study, split into 4 phases. Pick your weekly pace and the timeline adjusts.

Estimated finish: 68 weeks

What the exam covers

Source: official exam guide. Weights can change, so check before you start.

Before you start

• Admission requirements: a conferred bachelor's degree with an undergraduate GPA of 3.0 or better
• College calculus and linear algebra (multivariate derivatives, matrix/vector notation), probability (distributions, expectation, independence) and solid programming in Python, C/C++ or similar
• Tuition is about $21,000–$26,000 in total. Each course runs 10 weeks at roughly 10–25 hours per week

1. Weeks 1–6 · 60 h

   Maths and programming readiness

   • Compute gradients of vector expressions (for example the gradient of a least-squares loss) and work with eigendecompositions
   • Solve probability problems on conditional probability, Bayes' rule, MLE and Gaussian distributions
   • Write vectorised NumPy and basic PyTorch code comfortably

   Hands-on: Do the CS229 problem set 0 style review (linear algebra and probability). If any question takes more than 20 minutes, schedule extra maths review before you enrol.

   Use: CS229 Autumn 2018 lectures (Andrew Ng, YouTube) · MIT OCW 18.06 Linear Algebra

2. Weeks 7–23 · 170 h

   Required course: CS229 Machine Learning

   • Derive and implement linear and logistic regression, GLMs, generative models (GDA, Naive Bayes), SVMs and kernels
   • Implement neural networks with backpropagation, EM for mixture models, PCA/ICA, and the basics of RL
   • Complete the problem sets and a quarter-long final project, with a grade of B or better

   Hands-on: For your CS229 project, take a real dataset from your job or Kaggle, frame a clear prediction task and compare at least three model families with proper validation. Write it up in the course's conference-paper format.

   Use: CS229 course site

3. Weeks 24–38 · 150 h

   Second core or foundation: CS221 Principles and Techniques

   • Model problems as search, MDPs, games, constraint-satisfaction problems, Bayesian networks or logic, and implement solvers for each
   • Complete the programming assignments (for example search, MDP value iteration, game playing) to the B-or-better standard

   Use: CS221 course site · CS221 Autumn 2021 lectures (YouTube)

4. Weeks 39–68 · 300 h

   Electives: CS224N plus one more

   • In CS224N, build word vectors, a neural dependency parser and a transformer-based model, and complete a custom or default final project
   • In your second elective (CS231N, CS234 or CS336, for example), finish every assignment and the final project with a grade of B or better
   • Finish all four courses within three academic years

   Hands-on: For the CS224N final project, fine-tune or train a transformer on a task you care about. Report baselines, ablations and error analysis, and publish the code.

   Use: CS224N course site · CS224N lectures (YouTube) · CS336 Language Modeling from Scratch · Stanford Online YouTube channel

You're ready to book when…

• You can do a free past CS229 problem set from the YouTube course without help before you enrol
• Each completed course shows a grade of B (3.0) or better
• All four courses (at least one required and the rest electives) are finished within three academic years

Exam tips

• The certificate needs a B (3.0) or better in every course, not just a passing grade, so take one course per quarter if you work full time.
• Most students finish in 1–2 years. Electives are offered only in certain quarters, so plan your sequence from the course schedule before you start.