Keynotes
Data Mining for Ubiquitous Computing
Susanna Pirttikangas
Traditional data mining deals with large amounts of data that are not collected for some specific purpose. The analysis of data has usually been performed offline, and the aim has been to establish nontrivial and valuable information from vast amounts of digital data. The aims have not changed but while moving towards an ubiquitous world several challenges occur as the mining must be performed in mobile or embedded devices and in real time. There are several questions unanswered; Who performs the analysis? How to present the results in a (possibly) small screen? How to reduce computational complexity of the algorithms? Who has access to the information? Where is the data saved? How to process data streams? This tutorial focuses on different issues of ubiquitous data mining. The procedures required to perform data mining and context recognition are pointed out - from planning the testing environment to signal preprocessing, feature selection, pattern recognition and visualization of data. Furthermore, useful tools for performing the analyses are mentioned and future directions are given.
Bio:
Susanna Pirttikangas is a Researcher in Intelligent Systems Group (ISG)
at the University of Oulu. She finished her Master's studies at the
Department of Mathematical Sciences (1998) and received her PhD in
Electrical Engineering from the University of Oulu, Finland
(2004). She has worked as a visiting researcher in the Distributed and
Ubiquitous computing laboratory at Waseda University (2005-2006),
where her topic was small sensor nodes for activity recognition.
Her research interests include pattern recognition, data mining and
context recognition from wearable and environmental sensors.
SenseWeb: A Shared Infrastructure for Internet-scale Sensing
Aman Kansal
This talk discusses SenseWeb, a shared infrastructure tailored to run multiple simultaneous sensing applications on large collections of Internet-connected sensors operated by multiple entities. Compared to a stand-alone sensor network, such a shared infrastructure enables better resource re-use, extended spatial coverage, and reduced deployment costs. The talk discusses the requirements for effective sensor sharing and specifically the design of SenseWeb. Techniques are discussed for tasking large collections of shared sensors so that they can serve multiple concurrent applications while collecting data from heterogenous sensors with varying resource capabilities or willingness to share, different computational platforms, varying mobility and intermittent connectivity. SenseWeb provides a database abstraction that hides the messy details of data collection from applications, transparently caches raw and summarized sensor data, and uses a number of novel techniques to efficiently support multi-resolution spatio-temporal queries on real-time sensor data. The current prototype of SenseWeb has been publicly available for over an year and incorporates a large heterogenous collection of sensors including weather sensors, traffic sensors, parking sensors, and web cameras.
Bio:
Aman Kansal is a Researcher in the Networked Embedded Computing group
at Microsoft Research. He received his PhD in Electrical Engineering
from University of California Los Angeles in 2006. He received his MS
in COmmunications and Signal Processing (2002) and BS in Electrical
Engineering (2001) from Indian Institute of Technology Bombay. His
research interests include sensor-actuator networks, pervasive
computing, wireless and mobile computing, personal area networks,
Internet telephony, and embedded systems. Most recently, his work has
encompassed shared sensor networks of mobile phones, webcams, and
other sensors, controlled mobility in sensor networks, energy
harvesting theory and systems, and embedded flash storage.