XAI618: Agent Artificial Intelligence

Information

Outline

Agent learning can be defined as learning from experiences for AI agent to perform optimal actions given observations and involves reinforcement learning as a core component. On top of reinforcement learning, it generally combines sequential modeling and world modeling. This course will first review the basic reinforcement learning and then focus on recent distributed deep reinforcement learning for large-scale agent learning. Throughout the course, a number of representative agent learning problems will also be introduced and how these problems can be solved using recent machine learning algorithms will be described. In addition, recent approaches for generalizable agent learning based on multi-task / multi-modal learning, self-supervised representation learning, world modeling, meta learning, and offline reinforcement learning will be introduced and discussed. Furthermore, recent embodied intelligence based on large language models for robotics will also be discussed.

Objective

Throughout the course, students will learn the backgrounds of recent agent learning and understand the key factors in cutting-edge algorithms for generalist AI agent. In particular, the course will cover the current trend based on large-scale agent learning and large language models for AI agent. Eventually, the course aims students to have enough knowledge and moreover an insight for the corresponding researches.

Pre-requisites

Basic knowledges in calculus, linear algebra, probability, machine learning, and deep learning are required.

Material

Lecture notes will be the main material of the course, and these do not come from a single textbook.

Grading

Schedule