UCL-MSc-Machine-Learning-Coursework

My PDF solutions to our courseworks for the 2024-2025 year

Grades

• COMP0086 Probabilistic and Unsupervised Learning : 100%
• COMP0085 Approximate Inference and Learning in Probabilistic Models : 100%
• COMP0083 Advanced Topics in Machine Learning (Convex Optimisation) : 99%
• COMP0083 Advanced Topics in Machine Learning (Kernel Methods) : 76%

Feedback

COMP0086 Probabilistic and Unsupervised Learning

Raw mark 199 marks / 310 Q1 25 marks / 25 great work. Q2 15 marks / 15 great work. Q3 60 marks / 70 a. 5/5 b. 5/5 c. 10/10 d. 30/30 e. 10/10 f. 0 g. 0 -- not done Q4 5 marks / 35 (a) good answer, full marks (5 marks); 4 (b) and 4 (c) not provided. Q5 59 marks / 70 (a) -2: derive the MLE by differentiation subject to constraint. (b) -3: are they independent? (c) -1: state the one-step transition probability (d) -5: correct implementation, but missing intelligent initialisation to encourage convergence (e)(f) good Q6 0 marks / 60 Q6 was not attempted. Q7 15 marks / 15 great work. Q8 20 marks / 20 Excellent solution. 8 (a): 6/6, (b): 9/9, (c): 5/5. TOTAL: 20/20

COMP0085 Approximate Inference and Learning in Probabilistic Models

Overall mark 100% Raw mark 222 marks / 265 Q1 40 marks / 40 great work. Q2 47 marks / 35 Q2: (a) Very good answer, but log scaling of posterior would have been beneficial in terms of simplification and numerical stability: 9/10. b): Very good answer: 5/5. (c): Good solution, but additional code comments would have been great. In addition, you didn't account for numerical stability by adding a small positive value to the covariance matrix (le.g. 1e-5 * I):7/10. (d) Very good answer: 5/5. e) Very good answer: 5/5. (f) Very good answer: 15/15 g) Good answer, but a more concrete mathematical formulation is missing: 3/5. TOTAL: 29/35 + 18 = 47 Q3 60 marks / 70 (a) good (b) -2: unclear explanation. there is no regularisation here, state which variables are the regressors. (c-d) good. (e) -2: where are the suspected latent features? (f) -3: there appears to be numerical errors in the free energy. Learned feature 5 is incorrect. (g) -3: partial explanation Q4 0 marks / 40 no solution Q5 30 marks / 30 a. 5/5 b. 10/10 c. 5/5 d. 10/10 Q6 45 marks / 50 The free energy values are not right

COMP0083 Advanced Topics in Machine Learning (Convex Optimisation)

It was not specified why the fista rate can be applied in 2.5 (-1 point).

COMP0083 Advanced Topics in Machine Learning (Kernel Methods)

The areas you fell down on were not deriving certain quantities which you were expected to (e.g. Regularized CCA, the approximate COCO - in full, and giving a formula for the f,g in approximate COCO).