Astronomy and Astrophysics (David van Dyk, organizer)


George Djorgovski (California Institute of Technology)
Virtual Astronomy, Information Technology, and the Revolution in Scientific Methodology


Thursday 2:00-2:30, Fountain II

Abstract:

All sciences, including astronomy, are now entering the era of information abundance. The exponentially increasing volume and complexity of modern data sets promises to transform the scientific practice, but also poses a number of common technological challenges. The Virtual Observatory concept is the astronomical community's response to these challenges: it aims to harness the progress in information technology (IT) in the service of astronomy, and at the same time provide a valuable testbed for IT and applied computer science (CS). Challenges broadly fall into two categories: data handling (or "farming"), including issues such as archives, intelligent storage, databases, interoperability, fast networks, etc., and data mining, data understanding, and knowledge discovery, which include issues such as automated clustering and classification, mutivariate correlation searches, pattern recognition, visualization in highly hyperdimensional parameter spaces, etc., as well as various applications of machine learning (ML) in these contexts. Such techniques are forming a methodological foundation for science with massive and complex data sets in general, and are likely to have a much broather impact on the modern society, commerce, information economy, security, etc. There is a powerful emerging synergy between the computationally enabled science and the science-driven computing, which will drive the progress in science, scholarship, and many other venues in the 21st century.



Christopher Genovese (Carnegie Mellon University)
Quasi-adaptive Confidence Bands and the Dark Energy Equation of State


Thursday 2:30-3:00, Fountain II

Abstract:

In nonparametric regression problems, estimators exist that adapt to unknown smoothness: that is, they attain essentially optimal rates of convergence to the true function simultaneously over a range of assumed parameter spaces. But for most scientific problems, a good estimator is by itself not enough for making effective inferences. Also needed is an assessment of uncertainty against which to gauge possible features of the unknown function. One appealing approach is to construct uniform confidence bands for the function, a pair of random functions (L,U) such that P_f{L <= f <= U} = 1 - a for some target confidence level a. Unfortunately, in many cases of practical interest, adaptation for confidence bands is either limited or impossible.

Here, I will describe an alternative notion of coverage that focuses on capturing the smooth structure of a function. I will develop procedures that yield adaptive confidence bands, where coverage is understood in this alternative sense, and show how such bands are used in practice. I will illustrate the approach and surrounding issues with an analysis of supernova data to make inferences about dark energy and the acceleration of the universe.


Robin Morris (NASA Ames Research Center)
Modern Statistical Methods for GLAST Event Analysis


Thursday 3:00-3:30, Fountain II

Abstract:

The Gamma-ray Large Area Space Telescope (GLAST) is scheduled for launch in 2007. The primary instrument on GLAST is the Large Area Telescope (LAT), designed to perform a full-sky survey in the 30MeV to 300GeV energy range. The observations made by this instrument will give deep insights into extremely high-energy events at cosmological distances.

The LAT is a pair-conversion instrument. It consists of a tracker - a four-by-four array of towers, where each tower consists of alternating conversion and detection layers. The conversion layers are tungsten foils, and the detection layers consist of silicon microstrips, with separate planes measuring x- and y- position. Beneath the towers is a segmented calorimeter. In this work we focus on the tracker, which is designed to estimate the direction of incident photons.

Incident gamma-rays convert to electron-positron pairs in the tungsten foils, and these charged particles trigger the silicon microstrips as they traverse the remaining layers. They also produce secondary charged particles, and secondary photons (which may undergo pair conversion later). The charged particles are scattered as they traverse each layer, losing energy and changing direction. To reconstruct the direction of the incident photon the energy split between the initial electron and positron must be well estimated.

The statistics of each simple interaction are well quantified. We describe work-in-progress on developing a fully statistical reconstruction methodology, that incorporates in detail the statistics of the simple interactions to compute the full pdf over the energy and direction of the incident photons. It uses model selection methods to estimate the probabilities of the possible geometrical configurations of the particles produced in the detector, and numerical marginalization over the energy loss and scattering angles at each layer.