Adapting the User Context in Realtime: Tailoring Online Machine Learning Algorithms to Ambient Computing
Key: SHRS08-1
Author: Johannes Schmitt, Matthias Hollick, Christoph Roos, Ralf Steinmetz
Date: October 2008
Kind: @article
Keywords: ambient computing, pervasive computing, context-aware computing, machine learning
Abstract: Ambient systems weave computing and communication aspects into everyday life. To provide self-adaptive services, it is necessary to acquire context information using sensors and to leverage the collected information for reasoning and classification of situations. To enable self-learning systems, we propose to depart from static rule-based decisions and first-order logic to define situations from basic context, but to build on machine-learning techniques. However, existing learning algorithms show substantial weaknesses if applied in highly dynamic environments, where we expect accurate decisions in realtime while the user is in-the-loop to give feedback to the system’s recommendations. To address ambient and pervasive computing environments, we propose the FLORA—multiple classification (FLORA-MC) online learning algorithm. In particular, we enhance the FLORA algorithm to allow for (1) multiple classification and (2) numerical input values, while improving its concept drift handling capabilities; thus, making it an excellent choice for use in the area of ambient computing. The multiple classification allows context-aware systems to differentiate between multiple categories instead of taking binary decisions. Support for numerical input values enables the processing of arbitrary sensor inputs beyond nominal data. To provide the capability of concept drift handling, we propose the use of an advanced window adjustment heuristic, which allows FLORA-MC to continuously adapt to the user’s behavior, even if her/his preferences change abruptly over time. In combination with the inherent characteristics of online learning algorithms, our scheme is very well suited for realtime application in the area of ambient and pervasive computing. We describe the design and implementation of FLORA-MC and evaluate its performance vs. state-of-the-art learning algorithms. We are able to show the superior performance of our algorithm with respect to reaction time and concept drift handling, while maintaining an excellent accuracy. Our implementation is available to the research community as a WEKA module.
View Full paper (PDF) | Download Full paper (PDF)
Official URL

The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a non-commercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, not withstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.