Linear Algebra

MIT 18.06 Linear Algebra (OCW, Spring 2010 listing), taught by Professor Gilbert Strang, is widely regarded as the single best free resource for learning the linear algebra that underpins machine learning, and it earns that reputation: the lectures move from matrix elimination to the Singular Value Decomposition with unusual clarity, and the OCW package includes 34 full lecture videos, problem sets with solutions, and past exams with solutions (the companion 18.06SC 'OCW Scholar' listing adds 36 short TA-led problem-solving videos). It is completely free and openly Creative-Commons licensed with no signup, but MIT OpenCourseWare offers no certificate, no grading, and no instructor support for self-learners. The course is genuinely rigorous undergraduate material (prerequisite: multivariable calculus, 18.02), not a quick ML-math primer, and several lectures near the end lean on differential equations (18.03) that the listed prerequisites under-disclose. It is the right pick if you want deep conceptual mastery and are willing to self-discipline; it is the wrong pick if you want a fast, applied, ML-specific path with a shareable credential.

Outstanding, free, authoritative teaching of the exact math ML depends on, but it is full undergraduate rigor with no certificate, no support, and no ML-specific framing, so it fits self-disciplined learners who want depth rather than people who need a fast applied credential.

Best for: Self-motivated learners who want a deep, durable understanding of linear algebra for machine learning, data science, or engineering, and who are comfortable studying from video lectures and problem sets without grading or support. Ideal for CS/ML students filling a math gap, self-taught practitioners who keep hitting eigenvalues/SVD/PCA without intuition, and anyone preparing for graduate-level ML courses (e.g. Stanford CS229) that assume this material.

Skip if: People who want a short, applied, ML-only crash course, a structured cohort with deadlines, or a shareable certificate. Also not ideal for true beginners weak in algebra/calculus (multivariable calculus is assumed), or for anyone who needs hands-on coding labs rather than chalkboard proofs and intuition (the course uses MATLAB only lightly on some problem sets).