CS 3450 - Deep Learning

3 lecture hours 2 lab hours 4 credits
Course Description
This course provides an in-depth introduction to the foundations of deep learning. Students will learn how to architect, train, and evaluate deep neural networks. Students will gain experience with backpropagation, a variety of network structures, and a variety of options for training networks. Practical applications will be covered such as healthcare, object recognition and tracking, natural language processing, and art. (prereq: CS 3400 )
Course Learning Outcomes
Upon successful completion of this course, the student will be able to:
  • Understand the process of backpropagation and its role in training deep networks
  • Compare options for training deep neural networks, for example, optimization methods, generalization methods, normalization, initialization methods
  • Apply transfer learning to leverage pre-trained models
  • Compare various modern network architectures, for example, convolutional neural networks (CNNs), encoder-decoders, generative adversarial networks (GANs), or transformers. Emphasis on topics can vary depending on the instructor's specialties
  • Evaluate training methods and alternative architectures based on model accuracy, constraints, and training performance
  • Apply deep neural networks in the context of real-world applications such as healthcare, object recognition and tracking, natural language processing, and art
  • Discuss the ethical implications of deep neural network applications

Prerequisites by Topic
  • An ability to train a machine learning model on a provided dataset using a framework such as Keras or Pytorch
  • An ability to assess the quality of a machine learning model
  • Understand the concepts and application of supervised and unsupervised learning
  • Understand the role of optimization in machine learning
  • Understand overfitting
  • Be familiar with basic linear algebra such as matrix multiplication
  • Be familiar with multivariate calculus such as partial derivatives

Coordinator
Dr. Josiah Yoder


Print-Friendly Page (opens a new window)