Tutorials
Two two-hour tutorials will be held on Sunday 23, from about 2pm to 7pm. Please find below information related to each of the tutorials.
1) An Introduction to Data Mining in Astronomy
Speaker: Sabine McConnell
Sunday, September 23, 2007
Abstract
Over the past decade, data mining has gained a solid foothold in a variety of research areas. Data mining approaches attempt to extract novel and potentially useful information from large datasets in a semi-automated manner, and combines traditional analysis methods from the field of statistics with algorithms from machine learning, databases technology and visualization techniques. Data-mining approaches are divided into two main categories: predictive and descriptive. Predictive, or supervised, data-mining techniques build models to forecast the value of unknown or future attribute values based on known feature values. Depending on the type of the target, or class attribute, predictive approaches are divided into regression (continuous target attribute) or classification (discrete target attribute) tasks. In contrast, descriptive approaches attempt to discover underlying structures in the data without prior knowledge of the type or value of the target attribute. Other applications of data-mining techniques include visualization of data, association rule mining, and outlier detection. To date, data-mining techniques have been successfully applied to a large range of astronomical problems such as the separation of stars and galaxies, classification of planetary nebulae, galaxies and stars, antimatter search in cosmic rays, detection of expanding HI shells and selection of quasar candidates. However, the characteristics of astronomical data such as the noise associated with the collection process, the existence of multiple measurements of the same object, or the size of the datasets, complicate the data-mining process and warrant additional care when applying data-mining techniques in this domain.
About the main Author:
Sabine McConnell is an assistant professor in the Department of Computer Science at Trent University, Peterborough. Her research interests include data mining, analysis and visualization of astronomical data, and distributed systems. She is also interested in the development and optimization of astrophysical simulation and modeling code, and the integration of data-mining approaches into the Virtual Observatory.
2) 3D Visualization in Astronomy
Speakers: Jens Kauffmann, Michelle Borkin, Michael Halle, Douglas Alan, Nick Holliman, Ugo Becciani, Claudio Gheller
Sunday, September 23, 2007
Abstract
The complexity and data size of multi-dimensional datasets from astronomical observations and simulations is growing at an accelerating rate. Exploring these data sets, discovering new structures, and analyzing them is increasingly difficult for astronomers and cosmologists using only traditional software programs based on two-dimensional image display. While "slice by slice" viewing of datacubes will always be part of the astronomer's computational toolkit, multi-dimensional data of complex or unknown structure is often more naturally explored using three-dimensional visualization tools. However, despite an growing need for 3D visualization software, only a small number of such tools are available to the astronomy community at this time. This tutorial will provide a foundation for understanding the opportunities that 3D data visualization offer to astronomy. Specifically, we will demonstrate examples of the use of 3D visualization in astronomy, introduce the computer graphics and visualization concepts relevant to astronomy visualization tools, and outline the development details and capabilities of several 3D visualization software projects currently available. We will also demonstrate the potential value of astronomical data exploration using stereoscopic displays. Ample time will be reserved to discuss the community's requirements for 3D visualization in astronomy. We hope that this tutorial contributes to the creation of a community of people interested in developing and using new methods for exploring multi-dimensional data sets.