Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userRobust Machine Learning Applied to Terascale Astronomical Datasets
119
0
0.0
(
0
)
Authors
Nicholas M. Ball
Ask ChatGPT about the research
No Arabic abstract
We present recent results from the LCDM (Laboratory for Cosmological Data Mining; http://lcdm.astro.uiuc.edu) collaboration between UIUC Astronomy and NCSA to deploy supercomputing cluster resources and machine learning algorithms for the mining of terascale astronomical datasets. This is a novel application in the field of astronomy, because we are using such resources for data mining, and not just performing simulations. Via a modified implementation of the NCSA cyberenvironment Data-to-Knowledge, we are able to provide improved classifications for over 100 million stars and galaxies in the Sloan Digital Sky Survey, improved distance measures, and a full exploitation of the simple but powerful k-nearest neighbor algorithm. A driving principle of this work is that our methods should be extensible from current terascale datasets to upcoming petascale datasets and beyond. We discuss issues encountered to-date, and further issues for the transition to petascale. In particular, disk I/O will become a major limiting factor unless the necessary infrastructure is implemented.
rate research
Read More
156 -
Nicholas M. Ball
, Robert J. Brunner
, Adam D. Myers (University ofn Illinois at Urbana-Champaign
2007
We present recent results from the Laboratory for Cosmological Data Mining (http://lcdm.astro.uiuc.edu) at the National Center for Supercomputing Applications (NCSA) to provide robust classifications and photometric redshifts for objects in the terascale-class Sloan Digital Sky Survey (SDSS). Through a combination of machine learning in the form of decision trees, k-nearest neighbor, and genetic algorithms, the use of supercomputing resources at NCSA, and the cyberenvironment Data-to-Knowledge, we are able to provide improved classifications for over 100 million objects in the SDSS, improved photometric redshifts, and a full exploitation of the powerful k-nearest neighbor algorithm. This work is the first to apply the full power of these algorithms to contemporary terascale astronomical datasets, and the improvement over existing results is demonstrable. We discuss issues that we have encountered in dealing with data on the terascale, and possible solutions that can be implemented to deal with upcoming petascale datasets.
We apply instance-based machine learning in the form of a k-nearest neighbor algorithm to the task of estimating photometric redshifts for 55,746 objects spectroscopically classified as quasars in the Fifth Data Release of the Sloan Digital Sky Survey. We compare the results obtained to those from an empirical color-redshift relation (CZR). In contrast to previously published results using CZRs, we find that the instance-based photometric redshifts are assigned with no regions of catastrophic failure. Remaining outliers are simply scattered about the ideal relation, in a similar manner to the pattern seen in the optical for normal galaxies at redshifts z < ~1. The instance-based algorithm is trained on a representative sample of the data and pseudo-blind-tested on the remaining unseen data. The variance between the photometric and spectroscopic redshifts is sigma^2 = 0.123 +/- 0.002 (compared to sigma^2 = 0.265 +/- 0.006 for the CZR), and 54.9 +/- 0.7%, 73.3 +/- 0.6%, and 80.7 +/- 0.3% of the objects are within delta z < 0.1, 0.2, and 0.3 respectively. We also match our sample to the Second Data Release of the Galaxy Evolution Explorer legacy data and the resulting 7,642 objects show a further improvement, giving a variance of sigma^2 = 0.054 +/- 0.005, and 70.8 +/- 1.2%, 85.8 +/- 1.0%, and 90.8 +/- 0.7% of objects within delta z < 0.1, 0.2, and 0.3. We show that the improvement is indeed due to the extra information provided by GALEX, by training on the same dataset using purely SDSS photometry, which has a variance of sigma^2 = 0.090 +/- 0.007. Each set of results represents a realistic standard for application to further datasets for which the spectra are representative.
We provide classifications for all 143 million non-repeat photometric objects in the Third Data Release of the Sloan Digital Sky Survey (SDSS) using decision trees trained on 477,068 objects with SDSS spectroscopic data. We demonstrate that these star/galaxy classifications are expected to be reliable for approximately 22 million objects with r < ~20. The general machine learning environment Data-to-Knowledge and supercomputing resources enabled extensive investigation of the decision tree parameter space. This work presents the first public release of objects classified in this way for an entire SDSS data release. The objects are classified as either galaxy, star or nsng (neither star nor galaxy), with an associated probability for each class. To demonstrate how to effectively make use of these classifications, we perform several important tests. First, we detail selection criteria within the probability space defined by the three classes to extract samples of stars and galaxies to a given completeness and efficiency. Second, we investigate the efficacy of the classifications and the effect of extrapolating from the spectroscopic regime by performing blind tests on objects in the SDSS, 2dF Galaxy Redshift and 2dF QSO Redshift (2QZ) surveys. Given the photometric limits of our spectroscopic training data, we effectively begin to extrapolate past our star-galaxy training set at r ~ 18. By comparing the number counts of our training sample with the classified sources, however, we find that our efficiencies appear to remain robust to r ~ 20. As a result, we expect our classifications to be accurate for 900,000 galaxies and 6.7 million stars, and remain robust via extrapolation for a total of 8.0 million galaxies and 13.9 million stars. [Abridged]
We apply machine learning in the form of a nearest neighbor instance-based algorithm (NN) to generate full photometric redshift probability density functions (PDFs) for objects in the Fifth Data Release of the Sloan Digital Sky Survey (SDSS DR5). We use a conceptually simple but novel application of NN to generate the PDFs - perturbing the object colors by their measurement error - and using the resulting instances of nearest neighbor distributions to generate numerous individual redshifts. When the redshifts are compared to existing SDSS spectroscopic data, we find that the mean value of each PDF has a dispersion between the photometric and spectroscopic redshift consistent with other machine learning techniques, being sigma = 0.0207 +/- 0.0001 for main sample galaxies to r < 17.77 mag, sigma = 0.0243 +/- 0.0002 for luminous red galaxies to r < ~19.2 mag, and sigma = 0.343 +/- 0.005 for quasars to i < 20.3 mag. The PDFs allow the selection of subsets with improved statistics. For quasars, the improvement is dramatic: for those with a single peak in their probability distribution, the dispersion is reduced from 0.343 to sigma = 0.117 +/- 0.010, and the photometric redshift is within 0.3 of the spectroscopic redshift for 99.3 +/- 0.1% of the objects. Thus, for this optical quasar sample, we can virtually eliminate catastrophic photometric redshift estimates. In addition to the SDSS sample, we incorporate ultraviolet photometry from the Third Data Release of the Galaxy Evolution Explorer All-Sky Imaging Survey (GALEX AIS GR3) to create PDFs for objects seen in both surveys. For quasars, the increased coverage of the observed frame UV of the SED results in significant improvement over the full SDSS sample, with sigma = 0.234 +/- 0.010. We demonstrate that this improvement is genuine. [Abridged]
Broadening access to both computational and educational resources is critical to diffusing machine-learning (ML) innovation. However, today, most ML resources and experts are siloed in a few countries and organizations. In this paper, we describe our pedagogical approach to increasing access to applied ML through a massive open online course (MOOC) on Tiny Machine Learning (TinyML). We suggest that TinyML, ML on resource-constrained embedded devices, is an attractive means to widen access because TinyML both leverages low-cost and globally accessible hardware, and encourages the development of complete, self-contained applications, from data collection to deployment. To this end, a collaboration between academia (Harvard University) and industry (Google) produced a four-part MOOC that provides application-oriented instruction on how to develop solutions using TinyML. The series is openly available on the edX MOOC platform, has no prerequisites beyond basic programming, and is designed for learners from a global variety of backgrounds. It introduces pupils to real-world applications, ML algorithms, data-set engineering, and the ethical considerations of these technologies via hands-on programming and deployment of TinyML applications in both the cloud and their own microcontrollers. To facilitate continued learning, community building, and collaboration beyond the courses, we launched a standalone website, a forum, a chat, and an optional course-project competition. We also released the course materials publicly, hoping they will inspire the next generation of ML practitioners and educators and further broaden access to cutting-edge ML technologies.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
