Real-time Classification of Massive Datastreams

Date: Wednesday, February 25, 2009
Time: 2:00pm - 3:00pm
Location: Bldg. 50B, Room 4205

Speaker:
Josh Bloom
Department of Astronomy
University of California - Berkeley

Abstract:

The streaming of digital data at unprecedented rates presents unique
challenges to existing machine-learning classification schemes. The very
nature of the astrophysical phenomena of interest -- time-variable
sources -- demands rapid assimilation of new data into a dynamic
knowledge base. Creating and implementing new algorithms in a real-world
environment, where computational resources are limited and
classification statements are thus inherently imperfect, requires
significant innovation, drawing from expertise in machine learning,
astrophysics, high-performance computing, statistical methods, and
database theory. I will describe on-going work at Berkeley towards the
construction of a real-time classification framework to help us gain a
deep physical understanding of the dynamic universe.

Host of Seminar:

    Rollin Thomas