Use this URL to cite or link to this record in EThOS:
Title: Evolving intelligent systems for ubiquitous computing technologies
Author: Perez, Javier Andreu
Awarding Body: Lancaster University
Current Institution: Lancaster University
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Ubiquitous Computing (UC) is a new paradigm of research aiming at creating new autonomous systems that involve humans in practical real-life situations and applications. Human behaviour follows a stochastic process and raises a high level of uncertainty. In this context, Evolving Intelligent Systems (EIS) form a new machine learning · concept for developing adaptive algorithms that work on-line and are therefore reliable in real -time applications, with a very low complexity. Their adaptation properties make this type of algorithms very suitable to address the problems of stochastic behaviours and high uncertainty. In this thesis, the theory behind EIS is explained in detail, including fuzzy rule-based inference and densitybased models. In addition, key contributions that ease their application to UC systems are highlighted. Two types of evolving classifiers, eClassO and eClassl, are proposed as well as their simplified version called Simpl_eClass. Feature processing and novel dimensionally reduction methods are proposed when necessary. The purpose is this way to address the design and computational challenges of four key subfields of UC, namely Human Activity Recognition, Mobile Computing, Scene Recognition and Ubiquitous Robots. The first entails the recognition of human activities by using pervasive wearable sensors. The second implies the implementation of the proposed algorithm in a mobile platform to detect novelties inside a video sequence and recover pictorial memories. The third consists in an image processing experiment for classifying entire images into categorical classes. Finally, the fourth introduces the development of an autonomous robotic leader-follower platform. The model settings for each experiment are detailed. Successful performance values and pattern recognition rates were achieved for the different challenges addressed. For example, classification rates ranged between 71% and 80% in the online human activity recognition case, 80% for the scene categorization problem, and 98% for the leader status recognition. Unsupervised recognition of novelties was also successfully evaluated through a user validation test. The experiments showed in all cases very good times of response and resource awareness. It is concluded that the use of EIS with on-line computation and fuzzy logic inference offers valuable assets to be exploited in future UC developments.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available