|
|
|
New Methods in Data Mining for Time-Domain Astronomy
|
|
|
| Space missions and giant astronomical synoptic surveys and will soon monitor much of the sky regularly, detecting vast numbers of
interesting, variable astronomical objects. These objects will range from small solar system bodies to distant quasars and extra-solar
planets. The objective of this proposal is to develop the tools necessary for this new treasure trove. Specifically, this project will: 1)
Develop new, fast and scalable techniques to identify extra-solar planets. 2) Develop new, model based techniques to distinguish
transiting extra-solar planets and other, presently undiscovered "oddities" from variable stars. 3) Create data mining methods for
anomaly detection in time series data in order to enable discovery. 4) Preserve novelty detection when the volume of data becomes so
large that the scientists are "removed" from the data. The two "specific" problems are: (1) the discovery and characterization of
transiting extra-solar planets and (2) the classification of and outlier detection among variable stars. The proposed work will explore
the scientific analysis and modeling of massive datasets of light-curves (66 million light-curves now available, growing to 100 billion in a
decade) from a broad range of perspectives. The unique complications posed by the data in this domain will drive research for data
mining techniques to allow these analyses. The specific project goal is to enable qualitatively new science specifically to enable profound
insights into the astrophysics of extra-solar planets and variable stars. More generally, the proposed work will provide new data
mining modeling and analysis tools that scale well to very large datasets and are fast for "real time" analysis. This interdisciplinary
proposal will lead to the development of new insights, while training leading members of a powerful new generation of scientists. |
|
|
