ML Study Hub

Exam patterns and behaviour

What each paper looks like, which topics it leans on, and how much of it repeats. Built from the actual papers.

MLF

Machine Learning Foundations

Open MLF pattern

MLT

Machine Learning Techniques

Open MLT pattern

MLP

Machine Learning Practice

Open MLP pattern

The official structure

From the official IIT Madras BS course pages for MLF (BSCS2004), MLT (BSCS2007) and MLP (BSCS2008): each course has 12 weeks of coursework, weekly online assignments, 2 in-person invigilated quizzes and 1 in-person invigilated end-term exam. MLP lists MLF and MLT as prerequisites, and MLT lists MLF as a co-requisite.

The practice site groups papers by Quiz 1, Quiz 2 and End Term. We read what each group actually tests to infer the coverage:

ExamWhat the papers testCoverage
Quiz 1the first four weeks' material in all three coursesweeks 1 to 4
Quiz 2the next four weeksweeks 5 to 8
End Termeverything, with extra weight on the later weeksall 12 weeks
Tip

The week boundaries are inferred from the papers, not stated officially. Your own term's announcement of the quiz syllabus overrides this table. The grading weights (how much quizzes, end term and assignments count) are set by IIT Madras each term: check the current grading document on the official course page rather than trusting a number from a third party.

How the papers behave (patterns that hold across all three courses)

What the papers show

1. Papers come in parallel sets. One exam date is split across several sets of the same difficulty (forenoon and afternoon sessions, and separate sets for the foundation, diploma and improvement-attempt candidates). Names like QDF1/QDF2 (forenoon), QDF3/QDF4 (afternoon) and QDD (diploma) appear on the site. There is no single "exact set" to predict: all sets are drawn from the same topic pool.

What the papers show

2. Question types are fixed. Every paper mixes MCQ (one right option), MSQ (one or more right options) and SA (a numeric answer, sometimes with a range). Many papers add comprehension blocks: one passage with several questions under it. The tables above give each course's exact mix.

What the papers show

3. Duplicates live inside a date, types repeat across dates. All the parallel sets of one exam date share a pool of questions, so one date gives you many near-identical papers. Across different exam dates, exact repeats are rare (our measurement: a few percent at most). What does come back is the type: the same story with new numbers, the same trick in the answer options, the same comprehension shape. The "How much repeats" table on each course page shows both numbers.

What the papers show

4. The marks sit in a few repeating topics. Each course page lists the most-tested topics. If you can do those perfectly you have most of the marks before you touch the unusual questions.

What the papers show

5. Comprehension blocks repeat their structure almost word for word. The linear-programming story in MLF, the gradient-descent iteration table, the multivariate-normal block: learn one instance of each shape and the next instance is a different set of numbers.

A study loop that uses all this

  1. Learn the week from its lesson here (stories, solved examples, your-turn problems).
  2. Do that week's real-question page (linked from the lesson) to feel the exam wording.
  3. Once a week is done, sit a complete past paper under the clock: newest papers last, so you keep the freshest ones as honest mock tests.
  4. After each paper, write the reason each miss happened (concept, arithmetic, misread, rounding). That list is your private trap sheet.
  5. In the last two days: formula sheets, trap lists, and one full paper per course.

What the site cannot tell you