Publications

What is a Publication?
38 Publications visible to you, out of a total of 38

Abstract

Not specified

Authors: S. D. Kügler, K. Polsterer, M. Hoecker

Date Published: 1st Apr 2015

Publication Type: Journal

Abstract

Not specified

Authors: Maximilian Hoecker, Kai Lars Polsterer, Sven Dennis Kugler, Vincent Heuveline

Date Published: 2015

Publication Type: InProceedings

Abstract (Expand)

Regression tasks are common in astronomy, for instance, the estimation of the redshift or the metallicity of galaxies. Generating regression models, however, is often hindered by the heterogeneity of the available input catalogs, which leads to missing data and/or features of differing explanatory power. In this work, we show how simple but effective feature selection schemes from data mining can be used to significantly improve the performance of regression models for photometric redshift and metallicity estimation (even without any particular knowledge of the input parameters' physical properties). Our framework tests huge amounts of possible feature combinations. Since corresponding (single-core) implementations are computationally very demanding, we make use of the massive computational resources provided by nowadays graphics processing units to significantly reduce the overall runtime. This renders an exhaustive search possible, as we demonstrate in our experimental evaluation. We conclude the work by discussing further applications of our approach in the context of large-scale astronomical learning settings.

Authors: K. L. Polsterer, F. Gieseke, Christian Igel, Tomotsugu Goto

Date Published: 1st May 2014

Publication Type: InProceedings

Abstract (Expand)

Nearest neighbor models are among the most basic tools in machine learning, and recent work has demonstrated their effectiveness in the field of astronomy. The performance of these models crucially depends on the underlying metric, and in particular on the selection of a meaningful subset of informative features. The feature selection is task-dependent and usually very time-consuming. In this work, we propose an efficient parallel implementation of incremental feature selection for nearest neighbor models utilizing nowadays graphics processing units. Our framework provides significant computational speed-ups over its sequential single-core competitor of up to two orders of magnitude. We demonstrate the applicability of the overall scheme on one of the most challenging tasks in astronomy: redshift estimation for distant galaxies.

Authors: F. Gieseke, Kai L. Polsterer, Cosmin Eugen Oancea, Christian Igel

Date Published: 17th Mar 2014

Publication Type: InProceedings

Abstract (Expand)

Nowadays astronomical catalogs contain patterns of hundreds of millions of objects with data volumes in the terabyte range. Upcoming projects will gather such patterns for several billions of objects with peta- and exabytes of data. From a machine learning point of view, these settings often yield unsupervised, semi-supervised, or fully supervised tasks, with large training and huge test sets. Recent studies have demonstrated the effectiveness of prototype-based learning schemes such as simple nearest neighbor models. However, although being among the most computationally efficient methods for such settings (if implemented via spatial data structures), applying these models on all remaining patterns in a given catalog can easily take hours or even days. In this work, we investigate the practical effectiveness of GPU-based approaches to accelerate such nearest neighbor queries in this context. Our experiments indicate that carefully tuned implementations of spatial search structures for such multi-core devices can significantly reduce the practical runtime. This renders the resulting frameworks an important algorithmic tool for current and upcoming data analyses in astronomy.

Authors: Justin Heinermann, Oliver Kramer, Kai Lars Polsterer, Fabian Gieseke

Date Published: 2013

Publication Type: InProceedings

Abstract (Expand)

The task of estimating an object's redshift based on photometric data is one of the most important ones in astronomy. This is especially the case for quasars. Common approaches for this regression task are based on nearest neighbor search, template fitting schemes, or combinations of, e.g., clustering and regression techniques. As we show in this work, simple frameworks like k-nearest neighbor regression work extremely well if one considers the overall feature space (containing patterns of all objects with low, middle, and high redshifts). However, such methods naturally fail as soon as only very few or even no training patterns are given in the appropriate region of the feature space. In the literature, a wide range of other regression techniques can be found. Among the most popular ones are regularized regression schemes like ridge regression or support vector regression. In this work, we show that an out-of-the-box application of this type of schemes for the whole feature space is difficult due to the involved computational requirements and the specific properties of the data at hand. However, in contrast to nearest neighbor search schemes, such methods can be employed to extrapolate, i.e., they can be used to predict redshifts for patterns in new, unseen regions of the feature space.

Authors: F. Gieseke, Kai L. Polsterer, Peter-Christian Zinn

Date Published: 1st Sep 2012

Publication Type: InProceedings

Abstract

Not specified

Authors: R. J. Assef, K. D. Denney, C. S. Kochanek, B. M. Peterson, S. Kozłowski, N. Ageorges, R. S. Barrows, P. Buschkamp, M. Dietrich, E. Falco, C. Feiz, H. Gemperlein, A. Germeroth, C. J. Grier, R. Hofmann, M. Juette, R. Khan, M. Kilic, V. Knierim, W. Laun, R. Lederer, M. Lehmitz, R. Lenzen, U. Mall, K. K. Madsen, H. Mandel, P. Martini, S. Mathur, K. Mogren, P. Mueller, V. Naranjo, A. Pasquali, K. Polsterer, R. W. Pogge, A. Quirrenbach, W. Seifert, D. Stern, B. Shappee, C. Storz, J. Van Saders, P. Weiser, D. Zhang

Date Published: 1st Dec 2011

Publication Type: Journal

Powered by
(v.1.14.2)
Copyright © 2008 - 2023 The University of Manchester and HITS gGmbH