A Survey on Network Congestion Control with Reinforcement Learning

PhD Qualifying Examination


Title: "A Survey on Network Congestion Control with Reinforcement Learning"

by

Mr. Xudong LIAO


Abstract:

Internet congestion control remains an active field of research in both 
academia and industry. Classical TCP congestion control algorithms heavily rely 
on hand-crafted heuristics to perform congestion control by mapping predefined 
congestion signals to specific control actions. However, these algorithms have 
been notorious for performance degradation when their assumptions about 
congestion and packet-level events are violated. To address this problem, a 
recently evolved thread of research has provided us with a plethora of online 
learning or reinforcement learning enhanced congestion control approaches.

In this survey, we present an up-to-date and thorough introduction to the 
advances that utilize learning-based methods to control congestion. We first 
give the background of Internet congestion control and reinforcement learning. 
Then we center on several congestion control schemes powered by reinforcement 
learning and present their paradigms. We conclude by surfacing the drawbacks of 
existing learning-based congestion control schemes and providing future 
directions that potentially improve existing schemes.


Date:  			Thursday, 13 October 2022

Time:                  	2:00pm - 4:00pm

Venue:			Room 4472
 			lifts 25/26

Committee Members:	Prof. Kai Chen (Supervisor)
 			Prof. Gary Chan (Chairperson)
 			Dr. Brahim Bensaou
 			Dr. Qifeng Chen


**** ALL are Welcome ****