This page will be updated shortly before the midterm and final exams to reflect what we actually covered this semester.

Final exam

Reading Guide:

Listed below are the minimum things you should know. This is not an all-inclusive list, but you should at least be prepared to do these things:

Reinforcement Learning

  • Calculating Q-function and finding optimal policy
  • MDP, use of discounted reward
  • Sample: HW5

SVM, Multiclass classfication, and Kernel SVM

  • Sample: HW6
  • Focus on What kinds of functions K(·, ·) can correspond to some feature map φ?
  • How to calculate φ(x) from x?
  • How to calculate the weight-parameters (w/theta) given the decision function? You will find examples in the sample exams
  • Impact of offset, impact of C and slack variable
  • One vs all multiclass classification, loss for multiclass classification

Neural Networks

  • MLP – use, how number of layers affect func. approximation, hyperparameters
  • activation functions
  • CNN – use, what is convolution, what is filter and how are filters different and similar from MLP
  • RNN – use, problem of RNN, how are weights different and similar from MLP
  • Backpropagation for 2-layer MLP, and how is multi-layer MLP backprop. different from that

Bias - variance, Regularization, and Cross-validation

  • How is model complexity connected to bias, variance, and test error for different ML models?
    • For example, How variance changes with neighbor increase in KNN?
    • For example, How bias and variance changes with sigma in RBF SVM?
    • For example, How bias and variance changes with decision tree depth?
  • How does cross-validation help us to better generalize?

Ensembling

  • How does ensembling help in learning task to better generalize?
  • Effect of Bagging/ Random forest on bias-variance
  • Effect of Boosting on bias-variance
  • Sequential vs parallel training in bagging and boosting
  • adaboost algorithm steps
  • random forest algorithm steps
  • how does adaboost update weights and choose training examples for sequential training

PCA

  • Relationship between Principal componenets and explained variance
  • How much data is captured with each principal component?

There will be no sample questions posted. Focus on homeworks for format of descriptive questions. Check the midterm sample for questions with given scenario, for example, on bias, variance, etc.

Midterm exam

Reading Guide:

Listed below are the minimum things you should be prepared to do. This is not an all-inclusive list, but you should at least be prepared to do these things:

Logistic Regression, Linear Regression

  • What is the cost function and what is the log-likelihood?
  • How do you obtain the gradient descent update rule from cost function?
  • How do you get to log-likelihood from h(x)?
  • Why do we need 0-1 and perceptron loss?
  • What is minimizing least squares?

Multiclass classification

  • How is Multiclass classification connected to Logistic Regression (the general idea)
  • what is their loss (how do you train multiclass)?

Naive Bayes (NB)

  • How to estimate parameters for likelihood functions using bayes rule? What parameters needed to be calculated to obtain P(Y|X)?
  • How do you handle continuous and discrete X in NB?
  • How many independent parameters we need to estimate for calculation of joint probabilities?
  • how does NB assumption improves it?
  • what are the subtleties of Naive Bayes?
  • How to get log-likelihood from P(Dtheta) and how get MLE and MAP estimate of theta from it?

Bias - variance, and Cross-validation

  • How is model complexity connected to bias, variance, and test error?
  • How does L1 and L2 regularization affect classifiers?
  • How does cross-validation help us to better generalize?

Decision Tree

  • Run Simulation of a decision tree
  • When does overfitting happen? How to avoid overfitting in decision tree?
  • How does the decision boundary look like?

PCA

  • What are principal components? How do you find them?
  • How to get reduced dimension?
  • How do you #components?

K-means, KNN

  • What are the problems of KNN? How do you solve them?
  • What is hierarchical clustering? What does the evaluation metrics evaluate in clusters?
  • KNN decision boundaries
  • How variance changes with neighbor increase in KNN?
  • Advantage and Disadvantage

Sample Questions

I am providing a sample exam. Please be reminded that not all the topics in the questions have been convered in our classes. You can ignore them.