Exam Content Guide

Contents

Exam Content Guide#

Overview#

Our final exam format will be similar to the midterm exam. It is a big quiz covering all the topics we’ve learned after midterm exam. You’ll have 1 hour and 50 minutes to complete the exam.

Important Note: You need to use your laptop or computer to finish the exam.

  • Dimensionality Reduction:

    • Curse of Dimensionality

    • Feature extraction

      • Principal Component Analysis

      • Nonlinear Dimensionality Reduction Methods:

        • Kernel PCA

        • Stochastic Neighbor Embedding (SNE)

    • Feature Selection

      • Feature extraction vs. Feature selection

      • Filter method,Fisher score, Wrapper method, Embedded method

      • Bayesian Belief Networks, Markov Blanket & Boundary: Incremental Association Markov Blanket(IAMB)

  • Classification:

    • Decision Tree, Regression Tree, Information Gain, Gini impurity

    • Regression Tree

    • Overfitting and Tree Pruning

    • Bayesian Classification

      • Bayes’ Theorem

      • Classification Based on Bayes’ Theorem

      • Naïve Bayes Classifier

        • Avoiding the Zero-Probability Problem

    • Lazy vs. Eager Learning

      • Lazy Learner(Instance-Based Methods): k-Nearest Neighbor Algorithm

    • Linear, Logistic regression and Perceptron

    • Classifier Evaluation and Selection

      • Classifier Evaluation Metrics: Confusion Matrix, Accuracy, Error Rate,Sensitivity Specificity, Precision and Recall,and F-measures

      • Evaluating Classifier Accuracy: Holdout, Cross-Validation Methods, Bootstrap

      • Model Selection: ROC Curves

  • Clustering

    • What is Clustering? What is the difference between Clustering and Classification?

    • K-Means: Algorithm, Strength and Limitations

      • Strength: Simple and Efficient.

      • Weaknesses: 1. Clusters with different sizesand densities.2. Non-spherical clusters.3. Sensitive to initial centroids

      • How can we deal with the initialization problem?

    • Agglomerative Clustering

      • Merging (Linkage) Criteria: Single Linkage, Average Linkage, Complete Linkage: Smallest maximum distance

  • Neural Network

    • Feed Forward Network

      • Architecture

      • Activation function

      • Loss function

      • Regularization

      • Optimization: backpropagation, learning rate

    • RNN: Limitations of Traditional Neural Networks on Sequences data, Long-Term Dependencies

      • Long Short Term Memory Networks (LSTMs), Gated recurrent unit (GRU)

    • RNN with Attention Mechanisms

    • Transformer: Encoder-Decoder Architecture, Word Embedding

      • Self-Attention: Q, K, V, Softmax

      • Multi-Headed Self-Attention, Positional Encoding, Residual Connections, Layer Normalization

    • Convolutional Neural Networks

      • Convolution Layer: Output volume size? Number of parameters in Convolution Layer?

      • Pooling layer

        • Max Pooling, Avg Pooling

      • Fully connected layer

      • 1-D Convolution for Sequences data

Contents