Mathematical Aspects of Machine Learning

Jeremy Teitelbaum
University of Connecticut
Department of Mathematics
Math 5800, Spring Term 2020

Description: This is a project-based course in which students will explore both practical and mathematical problems arising in machine learning and data science. Possible topics include random walks for graph embeddings, optimization techniques, Monte Carlo methods, autoencoders, and linear and non-linear dimension reduction.

Course background

• Course Information Page
• Skills and Interests Inventory.

References and News

• References
• MNIST
• Simplified Git Workflow
• Python Programming References
• Google DataSets Announcement
• Using the UConn hpc cluster gpu nodes

“Syllabus”

Note regarding COVID-19: In light of the COVID-19 pandemic, the balance of this course will be conducted online. The final project will continue to be a github repo or website. The university has announced changes to the pass/fail regulations and has waived the drop deadline. This class is already being offered pass/fail.

Notes from the first two weeks:

• Week 1
• Week 2

We will continue our practice so far:

• Mondays: I will meet with each group for about 15 minutes during class just to make sure everyone is making progress.

• Wednesdays: Each group will present where they are to the class. The rules are that everyone in the group must speak.- Fridays: I will usually give a brief presentation on a topic in machine learning. So far I’ve discussed:

  • Bias and Variance
  • Polynomial Regression
  • Curse of Dimensionality
  • Precision and Recall
  • Principal Components Analysis
  • Neural Networks
  • Convolution html – slides – pdf
  • The Convolution of Gaussians is Gaussian
  • Gradient Descent – slides
  • Momentum
  • Bayesian Ideas