FEATURE:

 

ICPR2006

Track 4

Invited Talk1

Click here for Top of Page
Right Arrow: Next
Right Arrow: Previous
Newsletter

Challenges for Data Mining in Distributed Sensor Networks

 

By Virginio Cantoni

(Professor, University of Pavia, Italy)

Review by:

Adrian Ion
Vienna University of Technology

 

Besides presenting current and future challenges, Professor Virginio Cantoni gave a comprehensive overview of the field and its history. The objective of Sensor Networks (SN) is knowledge-gathering through systems that can answer questions that no single device could answer. Each contributor acts to maximize its contribution and adds detail/precision/speed to the answer. Each contributor has a role in the overall process and cooperates with the others to understand phenomena in situ and in real time.

 

A typical sensor node is a small battery-powered board, including a processing element (microprocessor), some measurement devices (visual, infra red, sonar, etc.), and a (wireless) communication device. In a typical network topology, sensor nodes communicate between themselves and with the base station, usually restricting such communication to the nodes in their “vicinity”. Evolution of network topology and distribution of processing, from the initial hierarchical structure to the current highly dynamic structure was presented.

 

Wireless Sensor Networks (WSN) will reach up to 10,000 nodes with an end-to-end reliability of more then 99.9%. Two technological issues include: data memory management, which is very critical due to the small amount usually installed; and radio frequency device communication, with its parameters like latency, positioning, heterogeneity, scalability, and self-organization. Another important set of issues are the ones related to the independence of the system when adapting to new situations. These include self-configuration, self-localization, self-optimization, self-awareness, self-healing, etc. The choice of the operating system running on the sensor nodes was also  discussed with the concrete example of TinyOS.

 

Data management is the task of collecting data from sensors, storing data in the network, and efficiently delivering data to the users. Basic features of data management include: trade-offs energy-efficiency, flexibility, robustness, and locality. There are two possible approaches when defining the data management strategy: distributed database (data gathering is formulated as a database retrieval problem) and agent system (sensors are agents interacting according to multi-agent paradigms). Mobile nodes pose further challenges and open possibilities for more complex in-route data mining and fusion.

 

Regarding hypothesis formation, the naive strategy of collecting all data in a central computing node with high computational power results in undesirable energy-costly transmissions. Decentralized and distributed algorithms are two common solutions.

 

After presenting the main characteristics regarding networks of cameras and the concrete example called “Eye society”, some applications of distributed sensor networks were presented: Field and Border control, Environmental Monitoring, Intelligent Health-Care Network, Wearable sensors, Networked Info-mechanical System, Urban sensing, and Network Deployment Blueprint at a Health Clinic in Chicago.

MICA2 Sensor mote hardware.