COMS 6998 Fall 2026: Continual Learning and Memory Models
Overview
This is an advanced seminar course that will focus on foundational ideas, recent work, and applications addressing the long-standing aim in machine learnings to develop models that can continuously improve and learn new tasks and abilities while retaining old ones. Students will read, present and discuss research papers as well as obtain experience developing a system in the course project. Evaluation will be based mainly on a project involving original research by the students, as well as presentations and participation in discussions.
The class will have a major project component.
Prerequisites
This course is designed to bring students to the current state of the art, so that ideally, their course projects can make a novel contribution. A previous course in neural networks/deep learning is required, such as COMS 4776 or the equivalent. Another course in relevant machine learning is strongly recommended, such as: COMS 4771 (Machine Learning); STCS 6261 (Foundations of Graphical Models); COMS 4775 (Causal Inference). Familiarity with linear algebra, basic multivariate calculus, the basics of probability, and programming skills is expected.
Where and When
- Fall 2026
- Instructor: Richard Zemel
- Teaching Assistants:
- Location:
- Time: Wednesdays, 2:10-4pm
- Instructor Office hours: Wednesdays 4-5pm, Uris 506
- Course Information: Syllabus
- [35%] Class presentations and participation. Marks based on: clear and succinct presentations; comments beyond paper; degree of difficulty; participation in class discussion.
- [10%] Project proposal
- [15%] Project presentations
- [40%] Project report and code
Course Structure
Aside from the first two lectures, the class meetings will consist of student presentations and discussions. In these a different group of students will present on a pair of related papers covering an aspect of these methods. The final two class meetings will be devoted to project presentations.
In-class discussion will center around understanding the strengths and weaknesses of these methods, their relationships, possible extensions, and experiments that might better illuminate their properties.
The hope is that these discussions will lead to actual research papers, or resources that will help others understand these approaches.
Grades will be based on: