Unit 11: Reinforcement Learning, Deep Learning and MultiAgent Learning

Commentary

This unit introduces two more machine learning tasks to reflect the recent trends in this field; i.e., reinforcement learning and multiagent learning. As opposed to supervised and nonsupervised learning, reinforcement learning methods learn from some form of feedback, such as reward or reinforcement. The rewards are based on the definition of Markov decision processes (MDPs), introduced previously. Multiagent learning, on the other hand, explores the techniques and applications of machine learning in a distributed AI environment involving multiple agents. Both of the methods have promising applications in social, economic, and engineering systems.

Unit Purpose

When you complete this unit, you will be able to

• Define concepts and tasks of reinforcement learning and multiagent learning.
• Discuss passive and active reinforcement learning, such as Q-learning and TD learning algorithms.
• Discuss new ideas and prototype applications of multiagent learning, and their adopted methods.

Section 1: Reinforcement learning
Section 2: Deep learning
Section 3: Multiagent learning

Readings

Supplemental Unit Readings

Books:

Sutton, R. S., and Barto, A. G. (1998). Reinforcement learning: An introduction. Cambridge, MA: MIT Press.

Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT press. Available at: https://www.deeplearningbook.org/

Papers:

Review the papers presented in Vohra, R., & Wellman, M. (eds.). (2007). Special Issue on Foundations of Multi-Agent Learning. Artificial Intelligence, 171(7), 363-452.

Activities

• Explore state-of-the-art applications of reinforcement learning and multiagent learning.
• Explore the recent AAAI and IJCAI conference paper titles to see if the AI community still cares about these two fields, summarize the main outcomes of the Abstracts or full papers. Post your summarization and references in the course conference.