Learning to Compete in Networked Systems

ECE Seminar: Learning to Compete in Networked Systems


Starts at: January 25, 2018 4:30 PM

Ends at: 6:00 PM

Location: Scaife Hall 125

Speaker: Dr. Carlee Joe-Wong

Affiliation: Carnegie Mellon University

Refreshments provided: Yes

Link to Abstract

Link to Video (1)

Details:

Abstract
Users in networked systems often compete with each other for resources, e.g., users may share resources on a cloud computing server. While competition for resources occurs in a large variety of settings, users’ optimal response to this competition depends on the particular resource being used and the method of resource sharing. In this talk, I will examine two examples of resource sharing in networked systems, focusing on how users can find their optimal strategies to compete with others. First, I will consider ridesharing systems in which taxi vehicles compete to serve passenger demands. After they drop off their passengers, taxi drivers must decide where to drive so as to maximize the probability of picking up a new customer. Traditionally, this dispatch problem has been solved in a centralized control context, but such an approach requires significant coordination across multiple taxis. In this work, we instead propose a distributed, deep reinforcement learning approach in which individual taxis can decide where to go without coordination. Simulations on a New York City taxi dataset indicate that, despite this lack of coordination, our approach significantly improves passenger waiting times compared to a centralized approach. I will then consider a scenario in which users compete for cloud computing resources using an auction. Their success at winning the auction then depends on (1) other users' bids, and (2) the cloud provider's decisions on which bids "win." We propose a means for users to optimize their bids based on past history, and show that this results in 90% savings on Amazon's EC2 cloud services.

Bio
Carlee Joe-Wong is an assistant professor in Electrical and Computer Engineering at Carnegie Mellon University. She received her A.B. degree (magna cum laude) in mathematics, and M.A. and Ph.D. degrees in applied and computational mathematics, from Princeton University in 2011, 2013, and 2016, respectively. Carlee is broadly interested in optimizing networked systems, including applications of machine learning and pricing in wireless, energy, and transportation networks. From 2013 to 2014, she was the Director of Advanced Research at DataMi, a startup she co-founded from her research on mobile data pricing. She received the INFORMS ISS Design Science Award in 2014 and the Best Paper Award at IEEE INFOCOM 2012.