Home / Exam patterns / MLT

Machine Learning Techniques: exam pattern and behaviour

Computed from every MLT paper on the practice site (171 papers, 3173 questions), plus the pattern notes from your earlier study sessions.

What a paper looks like

Exam	Papers	Years	Typical marks	Typical no. of questions	Share of questions	Share of marks	Papers with a comprehension block
Quiz 1 weeks 1-4	60	2022-2026	40	14	MCQ 33% / MSQ 21% / SA 46%	MCQ 32% / MSQ 24% / SA 44%	93%
Quiz 2 weeks 5-8	43	2022-2026	40	15	MCQ 39% / MSQ 15% / SA 46%	MCQ 39% / MSQ 16% / SA 45%	98%
End Term all 12 weeks	68	2022-2026	50	19	MCQ 35% / MSQ 22% / SA 43%	MCQ 32% / MSQ 27% / SA 41%	88%

Some uploads on the source site merge two parallel sets, which is why a few papers show double marks. MCQ = one right option, MSQ = one or more right options (strict marking), SA = type the number.

Which weeks the papers lean on

Week topic	Quiz 1	Quiz 2	End Term
1. PCA: finding the best view of your data	100%	10%	90%
2. Kernel PCA: curved views of data	98%	43%	76%
3. K-means: grouping without labels	100%	0%	100%
4. MLE, Bayes, Gaussian mixtures and EM	100%	100%	100%
5. Regression: least squares and its Bayesian view	10%	100%	98%
6. Ridge and LASSO: taming overfitting	2%	48%	73%
7. k-NN and decision trees	2%	100%	92%
8. Generative models: Naive Bayes	0%	86%	71%
9. Perceptron and logistic regression	10%	48%	100%
10. Support vector machines	2%	0%	100%
11. Bagging, boosting and AdaBoost	0%	0%	100%
12. Neural networks	0%	0%	80%

Share of papers whose concept summary mentions the week's topic (keyword match, so treat it as a guide, not a count of questions).

How much repeats

Exam	Papers	Exam dates	Question slots	Different questions	Copies of each (average)	Different questions seen on 2+ exam dates
Quiz 1	60	11	868	172	5.0	1 (1%)
Quiz 2	43	10	727	180	4.0	1 (1%)
End Term	68	13	1578	462	3.4	0 (0%)

Read this carefully. Within one exam date the parallel sets are drawn from the same pool, so each different question appears on several papers (the 'copies' column). Across different dates, questions with identical wording or identical figure files are rare. Matching here is exact, so a re-typed or re-cropped version counts as different; the real overlap in question type is much higher than this table can see. The takeaway: old papers teach you the types and stories that come back with new numbers, they are not a list of answers to memorise.

Questions that came back on a later exam date (exact match)

Exam dates it appeared on	Question (text-based ones)
2	Consider Lloyd’s algorithm used for k-means clustering and choose the correct statements:	open
2	Consider a dataset with 100 total data points. Each data point is classified as either type A or type B. We model this using a Bernoulli distribution, where p is the prob	open
2	Which of the following statements are true about the decision tree algorithm?	open
2	Do you have enough information to find the entropy of this node?	open

What each MLT paper rewards

MLT is the "algorithms" course, and the three exams split cleanly by week:

Quiz 1 (weeks 1 to 4): unsupervised learning and estimation. From the concept summaries: GMM in 52 of 60 papers, EM in 46, MLE in 39, Bayes in 32, PCA in 25, Bernoulli in 28, convergence in 22, covariance in 18, kernels in 15, the Beta distribution in 16, Beta posterior mean in 15, k-means++ in 10.
Quiz 2 (weeks 5 to 8): supervised learning I. (counts out of the 21 papers that have a summary) Decision trees (19 of 21), Naive Bayes (18), k-NN (12), MLE (12), MAP (11), information gain (11), linear regression (10), generative vs discriminative (9), kernel regression (8), ridge (7).
End Term (all weeks, heavy on 9 to 12). Soft-margin SVM (36 of 49 papers), hard-margin SVM (25), decision trees (24), logistic regression (23), AdaBoost (22), support vectors (20), ridge (20), overfitting (19), Naive Bayes (19), entropy (19), activation functions (19), neural networks (16), perceptron (13).

The questions that keep returning

What the papers show

Quiz 1. A Beta posterior mean after some heads and tails (the single most predictable calculation); one E-step responsibility from given likelihoods; one M-step mean or mixing weight; "does Lloyd's algorithm converge, and to what?"; k-means++ probabilities (squared distance); "which of these kernels are valid?" built from closure rules; variance explained or reconstruction error from eigenvalues; GMM parameter counts.

What the papers show

Quiz 2. Entropy and information gain from a small table; a Naive Bayes classification with Laplace smoothing; "what does a larger $k$ do in k-NN?"; ridge vs LASSO (which gives zeros?); "least squares equals MLE under what noise?" (Gaussian); LOOCV on three points.

What the papers show

End Term. Margin $=2/\|\mathbf w\|$ ; "which points are support vectors?"; the effect of a large $C$ (narrow margin); the SVM dual constraint $\sum\alpha_iy_i=0$ ; one AdaBoost update ( $\alpha_t$ and new weights); bagging vs boosting (variance vs bias); a perceptron update; the logistic gradient $(p-y)\mathbf x$ ; parameter counts for a small network; why identity activations add no power.

What this means for your preparation

Priority	Weeks	Why
Must be perfect	1, 4, 10	PCA, Beta posterior + EM, SVMs: the three biggest scoring areas
High	2, 3, 7, 8, 9, 11	kernels, k-means, trees, Naive Bayes, perceptron/logistic, boosting
Steady	5, 6, 12	regression, ridge/LASSO, neural networks

A fast win: MLT Quiz 1 is largely MLF Quiz 2 material applied to data. If you have done MLF weeks 3 to 6 (projection, eigenvalues, symmetric/PSD matrices, SVD/PCA), PCA and kernels are mostly vocabulary. That is the reason the weeks 1 to 4 braided plan exists.