|
April 2002
Bradley
Faculty member Elizabeth Bradley
has been awarded a Council on Research
and Creative Work (CRCW) grant-in-aid for "Feature Recognition in Oceanographic
Data". From Bradley's project description:
We are in the process of building a data analysis tool that can find and
recognize interesting coherent structures -- "features" -- in large scientific
datasets. There are two primary challenges in making such a tool both general
and practical; the definition of an interesting feature varies across domains,
as do the format and structure of the scientific data involved in identifying
one. Meteorologists, for instance, look for hurricanes in wind and pressure
data, while astrophysicists find supernovae by comparing telescope photographs,
and cell biologists recognize organelles using the density measurements that
are implicit in electron micrographs.
Intelligent analysis of any kind of data requires both domain-specific
principles -- e.g., what a hurricane looks like, and in what data -- and
general mathematics, such as the notion of a threshold or a gradient. Effective
automation of the data analysis process relies critically on good algorithms
that instantiate both kinds of ideas, and that work well on large amounts of
noisy data. For these reasons, the broad field that is variously termed data
mining, computer vision, pattern recognition, intelligent data analysis, etc.,
draws upon ideas and techniques from a very wide variety of disciplines,
ranging from statistics, geometry, and topology to digital signal processing
and the machine learning branch of artificial intelligence...
Scientific data -- the specific focus of our work -- has received surprisingly
little attention in these various research communities. Data mining, for
instance, is dominated by applications like detection of credit card fraud in
large databases; computer vision largely focuses on segmenting and
understanding the pixels in a camera image. The structure and formalisms of
science and engineering, however, can provide some serious leverage to feature
recognition techniques. Descriptions of the signature of credit card fraud in
a specific database are essentially impossible to construct a priori, but
scientific reasoning is presumably at least somewhat codifyable. This is useful
because data analysis tools work much better if they are provided with explicit
mathematical descriptions of the structures that they are to find.
The CRCW was created on October 1, 1935, to encourage and strengthen research
and creative work at the University of Colorado. The primary function of CRCW
is to provide faculty members with financial assistance and time free from
teaching responsibilities so they may pursue their research interests. CRCW
awards the Distinguished Research Lectureship, Faculty Fellowships, Junior
Faculty Development Awards, grants-in-aid, small grants, and conference support.
|
