Abstract

In recent years, technological advances have dramatically increased the quality and quantity of data available to astronomers. These new data streams are helping scientists make impressive strides in our understanding of the physical universe, but at the same time are generating massive data-analytic and data-mining challenges for the scientists who study the resulting data. This talk will discuss challenges that arise when modeling the physical processes that underlie two particular types of astronomical images.
 
In the first example, massive streams of multi-filter image of the Sun are used to study thermal structure in the solar corona. We develop an image segmentation framework that employs a parametric class of dissimilarity functions to reduce the data volume while preserving as much thermal information as possible for later downstream analyses. The result is a massive increase in the computational efficiency over existing methods, and, if preliminary results are substantiated, promising predictive power for thermal events.
 
In a second example, we develop formal statistical tests for unexpected structure that presents itself upon visual inspection of an image. To infer image structure, we conduct a Bayesian analysis of a full model using a multi-scale representation that allows flexible departures from the posited null model. As a test statistic, we use a tail probability of the posterior distribution under the full model. A novel aspect of our approach is that to reduce the computational demands of simulating under the null, we estimate an upper bound on a p-value, enabled by our choice of test statistic.
 
This is joint with with Nathan Stein, Vinay Kashyap, and Aneta Siemiginowska