Automatic Service Selection in Dynamic Wireless Network Environments
George Lee, Peyman Faratin, Steven Bauer, John T. Wroclawski · 2003
Our poster considers the use of machine learning agents to autonomously and continually select among wireless access services available to a user. Our context is the Personal Router project; a technical research program aimed at reshaping the wireless network access market towards greater competition and diversity [CW00]. By allowing potential suppliers to easily advertise service offerings in a given area, and allowing users to transparently and continually discover, negotiate terms for, and switch between available network services, we aim to greatly lower the barriers to entry in the wireless service market, creating a rich ecosystem of large and small service providers with different offerings, capabilities, and business models. A critical function within the PR vision is that of service selection. The Personal Router, acting on behalf of its user, must intelligently, intuitively, and autonomously select from among the available services the one that best meets its user’s needs. It must do this transparently, without involving, bothering, or distracting its user. Its task is difficult – acceptable service choice depends not only on the features and cost of the service, but also on the context of the situation, including such dynamic variables as the applications the user is running and the user’s higher level goals. Beyond this, the set of available services may change rapidly with time and the user’s location, requiring the PR to choose new services frequently. Without this automatic selection capability, users could not possibly respond to the constantly changing set of services and application demands. Previous approaches to wireless service selection, including static policies and manual selection, are inadequate. Static policies cannot accommodate individual user preferences, and users are reluctant to constantly interact with a user interface for manual selection, particularly in rapidly changing and complicated wireless service environments. These considerations motivate us to explore a machine learning approach, in which an intelligent agent learns user preferences and makes selections automatically on behalf of the user with minimal user involvement. Our poster presents this approach in three parts. First, we give an overview of the challenges, complexity, and importance of the service selection problem and the assumptions we make about the network and user. Next we present an architecture for service selection in the PR and describe its components. Finally we illustrate the performance of our system in dynamic and partially unobservable network environments.