[Colloq] Colloquium Talk - Bo Sheng - Tuesday, August 4, 10:30am

[Rachel Kalweit]

The College of Computer and Information Science presents a Colloquium Talk 

Speaker: Bo Sheng, College of William and Mary

Date: Tuesday, August 4
Time: 10:30am
Where: 366 West Village H

Title: Efficient Data Acquisition in Pervasive Computing
Environments

Abstract: Pervasive computing captures the vision that small,
inexpensive, robust networked processing devices are distributed at
all scales throughout everyday life. The large scale deployment of
these small devices (such as sensors and RFID tags) generates a lot
of data serving various applications. My major research goal is to
query such massive data sets in an efficient and secure manner.
Since these small devices have weak computation and communication
abilities, designing efficient and secure protocols is extremely
challenging, especially for complicated queries. We have developed
various algorithms for different data acquisition problems
exemplifying the design principles for those small devices. In this
talk, I will focus on our research on the efficiency issue and
present two recent work, finding popular items in an RFID system and
detecting outliers in a sensor network.

Finding popular items provides the information about the majority
which is useful in many RFID applications. Conventional solution of
collecting all data from every RFID tag is not efficient in terms of
scanning time, especially in a large scale RFID system. We propose a
novel randomized algorithm based on the idea of group testing which
allows us to quickly derive popular groups of tags. We demonstrate
that even though an RFID tag has extremely limited computation
ability, we can efficiently accomplish complicated tasks by
harnessing the power of randomization.

Outlier detection in sensor networks is another representative work
for efficient data acquisition. Outliers represent a complex form of
abnormal data as its definition involves the global information
about the data from many sensors. In a sensor network, however, it
is very challenging to efficiently obtain this global information.
We propose a histogram-based method for outlier detection. After
collecting and analyzing small-sized histograms, we can filter out a
lot of unnecessary data and identify potential outliers. Our
evaluation shows that this histogram method can dramatically reduce
the communication cost.

--------------------------------------------------------------------
Bio:

Bo Sheng is currently a Ph.D. candidate in Computer Science
Department at College of William and Mary. He received his B.S. in
Computer Science from Nanjing University, China. His major research
interests are wireless networks and embedded systems with emphasis
on efficiency and security problems.

Host: Guevara Noubir