One of the dominant areas of ubiquitous computing is a concept widely investigated nowadays: smart spaces. From the users’ perspective, the goal of a smart space is to enhance user experience by offering personalized context-dependent services and applications, which are beyond those that can be offered by a single device. This goal is almost impossible to achieve for all users and in all contexts by using applications that are preprogrammed to a fixed behavior. Therefore, smart space services and applications interact with their users and adapt their behaviors by using contextual information gathered from the physical environment (e.g. via sensors and user interaction) as well as information taken from online databases (e.g. from the internet via user profiles). The use of machine learning is particularly suitable for this purpose since a learning smart space application, that improves its performance in a given task with experience, is adaptive by definition. In this presentation, I will give an overview of the main practical challenges we have come across so far in applying machine learning in the setting of a smart space and present the solution approaches that we applied on a smart lighting test bed at the Intelligent Lighting Institute (ILI). In particular, we will look at the challenges in collecting data sets, the problem of missing data and how to deal with it, and the choice of a good performance metric for comparing different machine learning algorithms.

About Tanir Ozcelebi:

Tanir Ozcelebi received the B.Sc. degree in electrical and electronics engineering from Bilkent University, Ankara, in 2002 and the Ph.D. degree in electrical engineering from Koc University, Istanbul, in 2006. During Ph.D. studies, he developed multiple objective optimization models for content adaptive online streaming of below entertainment quality videos. In 2006, he joined the research group System Architecture and Networking (SAN) at Eindhoven University of Technology as a postdoctoral researcher. For two years his research activity was mainly on enhancing the Quality of Service for streamed multimedia in Next Generation Networks (NGN) and IP Multimedia Subsystem (IMS). Since 2008, he is an assistant professor at the SAN group. His current research interests involve resource and quality of service management for networked systems, machine learning and smart spaces. He is an active member of the Intelligent Lighting Institute (ILI) and leads the Adaptive Lighting Environments (ALE) project.