Interplay Between Reinforcement Learning and Generative Artificial Intelligence (Fall 2025)

Course Description

This course explores foundational research at the intersection of reinforcement learning (RL) and generative artificial intelligence (GenAI). This course has three main objectives:

• to introduce reinforcement learning for generative models such as language models (LMs) and diffusion models (DMs);

• to examine the use of generative models in reinforcement learning;

• to identify key limitations, open challenges, and promising research directions at this intersection.

The course emphasizes a balanced understanding of theoretical aspects and practical algorithms.

Learning Objectives

• Understand reinforcement learning methods for generative models, including language models and diffusion models.

• Examine the use of generative models in reinforcement learning and decision-making.

• Identify limitations, open challenges, and promising research directions at the intersection of RL and GenAI.

Prerequisites

A foundation in linear algebra, probability, calculus, and convex analysis is required. Familiarity with machine learning and/or reinforcement learning is recommended.

Requirements

• 10% of the grade: Class participation, including questions, discussions, scribing lectures, and providing feedback.

• 40% of the grade: Four homework assignments, each worth 10%.

• 50% of the grade: A project presenting and discussing a research paper, with 20% for the presentation and 30% for the report.

Communications

Canvas is used for course questions and submissions. Students can also discuss course material after class or arrange an appointment with the instructor.

Required Materials

There is no required textbook. Research papers are linked alongside the corresponding topics on the schedule. Lecture slides and problem sets are available for download from this site and on Canvas.