CS 445 -- Machine Learning

Final Exam Logistics

This exam is an open book, open notes, closed Internet exam. This exam is to be completed on your own with no collaboration with others. The exam, once started, needs to be completed within 2 hours. The exam questions will be supplied in Canvas. Anything that might be easier for you to draw can be drawn by hand (or on your computer) and attached to your Canvas submission.

You can take this exam anytime starting on Monday, December 14th. However, you must submit it by the end of our exam period, which is Tuesday, December 15th at 10:30 am. To make sure that this exam does not interfere with your other exams, it has been designed to be completed in under 2 hours (and thus, you could take it from between 8:30 am and 10:30 on the 15th, or earlier if you so choose).

Final Exam Material

  • Neural Networks and Convolutional Networks
    • Why are nonlinearities very important in multinode neural networks?
    • Why are the weights randomly initialized
    • Be able to compute the loss function for a single epoch
    • Explain the concepts of epoches and learning rates
    • Filters/Kernels in CNN
    • Discuss hyperparameters for all ANN and CNNs
    • Maxpooling layers -- explain their purpose and be able to compute the size of activations going in and out
  • Support Vector Machine (SVM)
    • Define the margin and how the C hyperparameter impacts the margin
    • Explain why this linear classifier can separate data that is not linearly separarable
    • Is this classifier type deterministic, or does it randomly initialize items and potentially converge to different solutions like ANNs
  • Dimensionality Reduction
    • Principal Component Analysis
    • Non-linear Embeddings
  • Class Imbalance
    • Undersampling and oversampling
    • Discuss oversamping and its how it effects the classifiers we have discussed in this class
    • SMOTE
  • Discuss ROC curves, their purpose and the information they convey
  • Clustering
    • Define and explain any portion of the K-Means algorithm. Be able to perform a step of k-means by hand
    • Selecting k -- how to investigate and select this hyperparameter
    • Define Hierarchital clustering and any/all of its steps. Be able to perform this type of clustering by hand
    • Be able to discuss any portion of the DBScan clustering method
    • Be able to length strengths/weaknesses of any of the clustering methods discussed
    • Define silhouette and the information it conveys