Self-Adapting Multi-sensor Systems: A Concept for Self-Improvement and Self-Healing Techniques

Abstract

Activity Recognition (AR) Systems more and more find their way into our daily lives, from monitoring daily activities to support in medical care. However, such systems tend to be used with narrowly defined specifications, demanding for application-dependent setup and configuration by their users. A long term goal are autonomous systems, being able to work with no (or minimal) user interaction. Closely related to that vision is the ability of autonomously adding further input sources (e.g., sensors) at run-time, leading to an increased dimensionality of the input-space. Our approach aims at systematically investigating methods necessary for the creation of self-adapting classification systems. This includes an architecture, based on Organic Computing (OC) principles, as well as the development of measures for comparing probabilistic models and procedures for evaluating classifiers of different dimensionality. With such evaluation techniques, systems should be able to adapt their system model at run-time in a self-organized manner. Besides self-improvement (adding a new sensor) we also address the problem of self-healing (replacing a sensor that dropped out).