IITM BS Machine Learning Foundations (BSCS2004): Syllabus and Tips

Machine Learning Foundations (BSCS2004), or MLF, is a 4 credit diploma level course in the IITM BS in Data Science. It is a fixed course in the Diploma in Data Science. It has no prerequisite of its own, but like every diploma course it opens only after you pass the 8 foundation courses. MLF is mostly maths: calculus, linear algebra, optimisation and probability, each shown through a machine learning problem.

The course page lists three instructors from IIT Madras: Harish Guruprasad Ramaswamy, Arun Rajkumar and Prashanth LA.

What you learn

The course page says MLF covers basics that are not machine learning themselves but are needed to understand it.

• Weeks 1 and 2: the setting. What makes a problem a machine learning problem, then a calculus review.
• Weeks 3 to 6: linear algebra. Least squares regression, eigenvalues and eigenvectors, symmetric matrices, and singular value decomposition. Week 6 uses principal component analysis on images.
• Weeks 7 to 9: optimisation. Unconstrained optimisation, convex sets and functions, then constrained optimisation with Lagrange multipliers. Logistic regression appears here as an optimisation problem.
• Weeks 10 to 12: probability models. Examples of probabilistic models in ML, the exponential family of distributions, parameter estimation and expectation maximisation.

The course page also lists the skills it wants you to gain. Two stand apart from pure maths: writing and debugging simple gradient descent code, and learning the parameters of Gaussian mixture models.

How it is assessed

The course page lists the standard pattern: 12 weeks of coursework, weekly online assignments, 2 in-person invigilated quizzes and 1 in-person invigilated end term exam. Check your term's grading document for the weights.

Where it counts

• Diploma in Data Science. MLF (4 credits) is one of the fixed courses every student must pass. See the Diploma in Data Science.
• Machine Learning Techniques. The MLT course page lists MLF as a corequisite.
• Machine Learning Practice. The MLP course page lists MLF as a prerequisite.
• Electronic Systems students. The ES handbook lists MLF (CS2004, 4 credits, no prerequisite) as an open elective and as the first course of the ES Minor in Machine Learning.
• Terms. The DS handbook says all diploma courses are offered every term.

The course page, the DS handbook and the ES handbook all give MLF 4 credits. The ES handbook also shows no prerequisite, the same as the course page.

Who finds it hard

If Mathematics 2 was a struggle, MLF will feel steep from week 3. Projections, eigenvalues and matrix factorisation come one after another. Weeks 7 to 9 add convexity and Lagrange multipliers, which may be new to you. Formula lists alone will not carry you through the quizzes. You need to have worked small examples yourself.

How to prepare

1. Revise Maths 2 before the term. Matrices, rank, projections and partial derivatives. Our guide to Mathematics for Data Science II lists what that course covers.
2. Work in 2 by 2. Try every new idea on a 2 by 2 matrix by hand before a larger one.
3. Code gradient descent early. Take f(x) = x squared, start at x = 1, and use a step size of 0.1. Each step multiplies x by 0.8, so x goes 1, 0.8, 0.64 and slowly reaches 0. Now try a step size of 1. The value jumps between 1 and minus 1 and never settles. This is one of the failure modes the course wants you to spot.
4. Link each topic to a use. Least squares goes with regression, eigenvectors with PCA, convexity with training. Write these pairs down.
5. Revise Stats 2 before week 10. Estimation and distributions return in the last three weeks. Our Statistics guide helps.

What to take before or after

MLF comes first in the ML sequence. You can take Machine Learning Techniques (BSCS2007) in the same term, since MLF is only a corequisite for it. Machine Learning Practice needs both done first. For how all three fit, read MLF, MLT and MLP explained.

Common questions

Should I take MLF and MLT together?

You are allowed to. If your Maths 2 is strong, taking both saves a term. If it is weak, take MLF first so MLT does not sit on shaky maths.

Is there coding in MLF?

Some. The course page expects you to program and debug simple gradient descent methods. Most of the course, though, is maths on paper.

Do I need to know eigenvalues before MLF?

No. Week 4 teaches eigenvalues and eigenvectors. But you should be quick with matrix multiplication and solving linear equations before then, or week 4 will feel rushed.