Surveys provide a major impetus in astronomy for understanding old and discovering new phenomena. With surveys we test hypotheses, improve statistics and attempt to put our understanding on a more secure basis or to extend it to new regimes. Through surveys we may also discover new types of objects, uncover new structures or observe new processes. Yet the survey alone is insufficient to achieve these: what we do with the data is decisive, and the techniques we use can make the difference between discovery and non-discovery (or false discovery). Analysing the large amounts of complex data delivered by current and future surveys is a major challenge and requires powerful techniques for their effective exploitation.
This workshop will focus on two important and closely related issues in data analysis. The first is the detection and classification of astrophysical objects or phenomena. In almost every survey we need to classify objects or estimate astrophysical parameters. Identifying and assembling samples of objects is a prerequiste to analysing their properties. How do we best use the data for classification? Do we want discrete classification or continuous parameter estimation? What parameters can we estimate and with what accuracy and reliability? How do we train our classifiers? The second theme is that of discovery. Not all objects can be classified into our pre-defined schemes, and "outliers" may mark new discoveries. Similarly, uncovering relationships between objects (e.g. via clustering or parameter correlations) is a route to learning about astrophysical phenomena. How do we maximise the chances of discovery (without an unmanageable level of false positives)? How do we best find structure in multidimensional data? How can we distinguish significant features from background noise or contamination?
These themes are linked by the increasing use in astronomy of machine learning or pattern recognition algorithms. Relevant methods are both supervised algorithms for classifying objects into pre-defined systems as well as unsupervised methods for detecting outliers, finding relationships between objects and discovering natural classes.
The purpose of this workshop is to bring together experts working on the analysis and interpretation of large, complex astronomical data sets. Many of the issues we face are common across different areas of astronomical research. These include;
Some of the techniques for addressing these are similar across fields or could be transferred to new problem domains. This workshop is an opportunity to learn about the latest work, to exchange ideas and to foster collaborations. Relevant areas of astrophysics which will be represented include