Doctoral Thesis DIS-2014-02

BibliographyFöll, Stefan: State-based context prediction in mobile systems.
University of Stuttgart, Faculty of Computer Science, Electrical Engineering, and Information Technology, Doctoral Thesis (2014).
193 pages, english.
CR-SchemaC.2.4 (Distributed Systems)
KeywordsContext Prediction; State-based Predictors; Markov Models; Context Update Protocols; Proactive Mobile Applications
Abstract

Context-aware computing has developed from a pure research area to a widely acknowledged design principle of modern mobile systems over the last years. Mobile applications, able to automatically adapt to a user’s dynamic context and improve the ease of human-computer interactions, are commonly available today. However, with current context-aware services such as restaurant finders or mobile tour guides, it is possible to support users with respect to their present behaviour only. As the next stage in context-aware computing more intelligent proactive applications are envisioned which can not only respond to the current, but also the future context of humans: Smart homes capable of controlling the ambient environment in expectation of the inhabitants’ prospective actions; Social network applications which alert users about places where their friends might be going to; Personalized mobile recommender system to promote events and offers at venues which are relevant to the daily schedules of humans.

The development of suitable context prediction methodologies to turn such applications into a reality is however a challenge. The reason is that future context information, hidden in the raw context traces left by users in the real world, is not immediately accessible to applications. Therefore, sophisticated context prediction approaches are required that are able to discover and mine patterns of a user’s behaviour from observed context histories. However, approaches which make accurate and expressive context predictions available and exploit this knowledge to optimize context-aware systems are missing in current research. As a consequence, the full potential of context-aware technologies has not been completely realised yet. In order to address this issue, we contribute in this work new context prediction algorithms and models for state-based context data, suitable for a range of different context types, such as a user’s locations or activities. To this end, this thesis makes the following contributions.

In the first part of this thesis, we develop a novel context prediction system which applies statistical modeling concepts to automatically learn a machine-processable model of a user’s behaviour and infer context predictions. With our context prediction system, we identify and address two shortcomings of existing approaches, prediction accuracy and prediction expressiveness, and propose suitable techniques and algorithms to improve them. For increasing the prediction accuracy over current systems, we develop a new context predictor that is able to exploit the conditional dependency of context changes on a user’s activities to anticipate forthcoming context states. Further, in order to overcome the limited expressiveness of prevailing prediction approaches, we explore the application of model checking algorithms for enabling expressive time-dependent forecasts in context prediction systems. Based on the algorithms and models developed in the first part of this thesis, we are able to significantly increase the amount and accuracy of the knowledge provided to proactive applications for the prediction of future context information.

In the second part of this thesis, we shift our attention towards tailored context prediction approaches to optimize the performance of mobile sensing applications. These applications represent a new class of mobile systems in the focus of current research, designed to forward streams of sensed context updates to interested parties over wireless communication channels. As mobile data communication induces a substantial energy overhead on mobile devices, we develop novel prediction-based protocols for improving the energy efficiency of mobile sensing applications. First, we present update protocols which are able to exploit context predictions for reducing the number of transmitted context update messages and trading off context accuracy vs. energy consumption. Then, we extend our approach and show how knowledge about a user’s future behaviour can be used to find the optimal update schedule for both sensing and communicating context data given hard bounds on the energy consumption on a mobile device.

We have implemented and validated our context prediction models in detailed experimental evaluations using synthetic and real-world context data. The results of our experiments demonstrate the effectiveness of our concepts for enhancing the accuracy and expressive power of predictions, as well as for increasing the energy efficiency of context-aware mobile systems.

Full text and
other links
PDF (6204035 Bytes)
Department(s)University of Stuttgart, Institute of Parallel and Distributed Systems, Distributed Systems
Superviser(s)Rothermel, Kurt; Ferscha, Alois; Diekert, Volker
Entry dateOctober 1, 2014
   Publ. Computer Science