Machine Learning with Graphs

Despite the catalog id, this is Stanford CS224W: Machine Learning with Graphs, taught by Professor Jure Leskovec (creator of node2vec and PyTorch Geometric's research lineage) — and it is the de facto standard graduate course for graph neural networks, with lecture videos, slides, and assignments published free at web.stanford.edu/class/cs224w. Independent self-study trackers (csdiy.wiki, cs61bbeyond) rate it 4/5 difficulty and estimate roughly 80 hours of work, and the recurring community consensus is that it is the single most-recommended GNN resource for people serious about the field. The free public materials cover node embeddings, the full GNN stack (message passing, GraphSAGE/GAT, GNN theory and expressiveness), heterogeneous and knowledge graphs, graph transformers, recommender systems, and generative models — but you only get YouTube lectures plus posted slides and ungraded Colab notebooks; there is no certificate, no grading, and no mentorship on the free track. Stanford separately sells a paid professional version (XCS224W via Stanford Online) that adds graded assignments and a certificate. This is a genuinely advanced course that assumes solid machine-learning fundamentals and Python, so it is the right pick for ML-literate engineers and researchers, and the wrong pick for beginners.

Outstanding, authoritative, and free for the right learner, but explicitly advanced: it presumes prior ML, linear algebra, probability, and Python/PyTorch fluency. Take it if you already know neural-network basics and specifically want graph ML; otherwise build foundations first.

Best for: ML-literate engineers, data scientists, and graduate students who already understand neural networks and want a rigorous, research-grade foundation in graph machine learning (node embeddings, GNNs, knowledge graphs, recommender systems). Ideal for self-learners comfortable reading papers and writing PyTorch/PyTorch Geometric code who want the most-cited free GNN curriculum rather than a gentle tutorial.

Skip if: Complete beginners or people new to deep learning. The course assumes machine-learning fundamentals, linear algebra, probability, and Python proficiency, and self-study trackers rate it 4/5 difficulty (~80 hours). Anyone wanting hand-holding, a quick certificate, or graded feedback on the free track should not start here — the free version has no grading, no mentorship, and no certificate.