Data-Driven Processing in Sensor Networks

                            Jun Yang
                        Duke University

Wireless sensor networks enable data collection from the physical
environment on unprecedented scales.  In this talk, I will describe
some data processing problems that arise in building an environmental
sensing network in Duke Forest, in collaboration with ecologists and
statisticians.  Because of severe resource constraints on
battery-powered sensor nodes, it is infeasible to collect and report
all raw readings for centralized processing.  An effective approach is
model-driven data acquisition, which avoids acquiring readings that
can be accurately predicted from known spatio-temporal models of data.
We argue for an alternative, data-driven approach, which exploits
models in optimizing push-based reporting, but does not depend on the
quality of models for correctness.  A particularly thorny issue with
push-based reporting is transmission failures, which are common in
sensor networks, and make failed reports indistinguishable from
intentionally suppressed ones.  The cost of implementing reliable
transmissions is prohibitively high.  We show how to inject
application-level redundancy in data reporting to enable efficient,
effective, and principled resolution of uncertainty in the missing
data.