summary.tex

%\centerline{\bf A Unifying Description of Astrophysical Variability for Event
%Classification} \medskip

Modern time-domain astronomical surveys are designed to uncover the wealth of
temporal astrophysical phenomenology: extrinsic, intrinsic, periodic, and
transient variability of all shapes and amplitudes.  Since of our understanding
of the underlying event physics comes from spectroscopic follow-up, the
rapid and effective classification of new imaging data, which feed spectroscopy, is of paramount importance. This urgency is felt today as follow-up resources are already saturated; the
ratio of interesting events to our ability to study them will only worsen
with next-generation alert streams. The challenge lies in separating the
prosaic from the novel in evolving data streams, using a holistic classification
infrastructure that spans all types of phenomenology. Since modern surveys will
be sampling events sparsely in time, and irregularly in wavelength, the optimal
variability models will be inherently temporal {\it and} spectral.  {\bf This
proposal will build upon the current state--of--the--art in event classification
to yield: 1) spectral--temporal models of astrophysical variability; 2) a new
taxonomy for classification of variables that is probabilistic and statistically
robust; 3) a broad range of techniques for classifying input data and an open
service that will enable these classifications for the community;  4) the
ability to probabilistically state what additional observations are required to
improve the classification}. Models will be built using photometric and
spectroscopic data from time-domain surveys of the past decade, fulfilling
their promise as precursor surveys that may be used to inform future efforts. As
well, theoretical priors will be used to build models for anticipated classes of
new variability.  These models will be applied to future real-time data streams
to classify events as they themselves evolve, and as our understanding of them
grows through additional photometric or spectroscopic sampling.   These
spectral--temporal surfaces are the logical culmination of the proposed event classification
efforts.

\bigskip \centerline{\ub{\sc Intellectual Merit of the Proposed Activity}}

This proposal will develop a classification infrastructure based upon the
unifying view that all astrophysical variability is an inherently spectral and
temporal process.  Such an effort is now possible due to the accumulation of
time-domain data over the past decades, and an increasingly more sophisticated
adoption of statistical techniques by astronomers. It is also necessary, given
the massive volumes of data expected from next-generation time-domain surveys. 
By deriving real-time knowledge from data, the proposed effort 
aims to help maximize the scientific return of (significant) NSF and national investments.
%The investigators of this proposal have an extensive history in the gathering of
%time-domain astronomical data, and in their real--time interpretation.  They
%are directly connected to past, current, and future time--domain surveys, and
%are thus ideally positioned to undertake this effort in a way that is guaranteed
%to be impactful.

\bigskip \centerline{\ub{\sc Broader Impacts of the Proposed Activity}}

The derived spectral--temporal models will be provided as a resource to the
astronomical community.  The proposed effort includes an automated classification resource
that accepts data from time-domain event streams, compares these evolving data
to the set of spectral--temporal models, and returns {\bf probabilistic}
classifications that an event is of a given type.  This resource will provide
value-added information to survey data, and will help the entire community to
sift through these increasingly more impenetrable event streams. This is an
inherently multidisciplinary effort, requiring astronomical domain knowledge to
aggregate the acquired data and classify known phenomena, and statistical tools
to build the spectral--temporal models and to use them in inference of an
incoming data stream. Students and postdocs will be trained in this multidisciplinary approach to a modern science challenge.