Title of Presentation: Segmentation Data Analysis
Primary (Corresponding) Author: Jeffrey D. Scargle
Organization of Primary Author:
Abstract: Much astrophysics, Earth and space science data are in the form of measurements distributed over a data space of some fixed dimension. Examples are:
time series data -- 1D
image data -- 2D
redshift galaxy surveys -- 3D
gamma-ray photon data -- 4D (space, time, energy)
The measurements can refer to a physical quantity measured over a pre-defined interval (e.g. pixel) or to discrete points distributed over the data space (e.g., X-ray or gamma-ray photon maps). In both cases, science analysis can be based on a density estimation procedure, followed by for example the detection of clusters or sources from the density map.
I have developed an algorithm that yields a segmentation analysis of 1D data, based on finding the optimal segmentation of the interval for a given fitness function (e.g. a likelihood for the piecewise constant model of the data). In addition, the optimal segmentation problem in a space of any dimension can be rigorously solved with an extension of the 1D algorithm.
Examples of such analysis include: