Practical Machine Learning

Dr. Suyong Eum / Dr. Hua Yang

home lectures assignments resources python workshop

Assignments

General

There will be three assignments during the semester. First and second ones will cover 30% of the total mark each, and the last third one will cover the rest 40%. Since there is not an examiniation in this subject, total marks of the course will be based on the assignments only.

ASSIGN#	Topic	Allocation	Due	Lecturer
1	Multi-class classificaiotn of English Words using Support Vector Machine (SVM)	30%	JUN. 8th	Dr.Eum
2	Image classification with Tiny imageNet using Convolutional Neural Network (CNN)	30%	JUL. 13th	Dr.Eum
3	Self-regulated open book examination	40%	AUG. 3rd	Dr.Eum

	Training data set for assignment 1 (HERE)
	Testing data set for assignment 2 (HERE)

Questions for assignment 3

#	Question	Topic
Q1	What are the differences between classification and clustering?	Week1
Q2	In which cases is the normal equation good for a linear regression problem (in terms of dimension & size of data)?	Week2
Q3	Why do we use different types of error functions: Mean Square Error and Cross Entropy Error?	Week2
Q4	Discuss the relation between mixing coefficient (𝛑) and responsibility (γ) in a GMM model.	Week3
Q5	Discuss the relation between GMM and HMM based on their corresponding parameters.	Week4
Q6	Why do we introduce the marginalization out (z_{k-1}) in the derivation of the solution for viterbi decoding algorithm?	Week4
Q7	What is the phytiscal meaning of lagrange multiplers shown in the dual problem of SVM?	Week5
Q8	Kernel trick: why is it called a trick?	Week5
Q9	What is the difference between X^TX and the covariance matrix from cov(X)?	Week6
Q10	Explain the relation between prinicipal components and eign values/vectors of the covariance matrix from cov(X).	Week6
Q11	Explain why cross entropy error function goes well with Softmax activation function.	Week7
Q12	Prove the partial derivative shown in slide 33 (it is related to Q11).	Week8
Q13	What do 1) epoch and 2) batch mean in the training of a machine learning model?	Week9
Q14	Explain the difference between gradient descent (GD) and stochastic gradient descent (SGD).	Week9
Q15	Why is LSTM robust against gradient vanishing problem compared to vanilla RNN?	Week10
Q16	Explain what dropout is and how it can be applied in RNN.	Week11
Q17	Discuss pros and cons of both VAE and GAN models.	Week12
Q18	What is Batch Normalization and why do we use it?	Week13
Q19	Explain why action value fucntion is preferable to state value function in RL.	Week14
Q20	How does experience replay help the convergence of DQN?	Week14
Q21	How does entropy regularization help the exploration in PG based RL algorithms?	Week15
Q22	Your comments on this course for its future improvement.	PML