Machine Learning in Community Sensing and Human Computation
There is a tremendous potential in harnessing sensing resources spread
throughout the population and infrastructure in order to achieve
benefits for society at large. In this series of talks I will
introduce two projects in Community Sensing. The Community Seismic
Network aims to rapidly detect and monitor earthquakes using
accelerometers in cell phones and other consumer devices. The
OpenSense project employs inexpensive sensors mounted on trams, buses
and bikes to provide high-resolution air quality maps, in order to
study health implications of exposure to air pollution. Both of these
projects share the goal of estimating complex phenomena (air quality,
earthquakes) based on large numbers of heterogeneous sensors operated
by the community. Motivated by these and other applications, e.g., in
Human Computation, citizen science and crowdsourcing, I will focus on
particular technical challenges such as: How can one piece together a
global picture from large numbers of noisy observations of a priori
unknown quality? How can one maximize the value of the information
gained under constraints such as communication, privacy, power and
budget? In particular, I will discuss how modern techniques from
machine learning, discrete optimization and probabilistic reasoning
offer solutions to these challenges.
Key references: