Basic course information

• Lecture times: Mondays 4:10-6:00pm
• Lecture venue: TBD
• Instructor: Daniel Hsu

Syllabus

(Additional details to be filled-in.)

Course goals and topics of study

The goal of this seminar course is to study the computational and statistical ideas underlying (or at least related/adjacent to) large language models.

Topics may include the following.

• N-gram and maximum entropy models
• Latent variable models
• Calibration
• Neural computation models
• Preference learning
• “In-context learning”
• “Chain-of-thought” learning

Readings from research papers and other sources will be made available on the course website.

The emphasis of this course is on theoretical/mathematical foundations, although the some of the motivations come from experiments and real systems.

Prerequisites

A course in computational learning theory or machine learning theory (e.g., COMS 4252, COMS 4773), theoretical statistics/neuroscience, or equivalent/similar technical preparation.