This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| GREAT10 | |
|---|---|
| Name | GREAT10 |
| Established | 2010 |
| Founders | International Astronomical Union; University College London; European Space Agency; STFC |
| Venue | virtual |
| Discipline | astronomy; cosmology |
GREAT10
GREAT10 was an international image-analysis challenge focused on weak gravitational lensing, organized to benchmark shape measurement algorithms against realistic simulated data. It convened a consortium of research groups and institutions including teams from University College London, ETH Zurich, Max Planck Society, European Space Agency, and observatories connected to projects such as Sloan Digital Sky Survey, Pan-STARRS, and Dark Energy Survey. The initiative aimed to quantify algorithmic biases relevant to forthcoming surveys like Euclid (spacecraft), Large Synoptic Survey Telescope, and Wide Field Infrared Survey Telescope.
GREAT10 grew from earlier community efforts that included the first GRavitational lEnsing Accuracy Testing challenge and parallel activities associated with collaborations at University of Edinburgh, University of Cambridge, and Imperial College London. Primary objectives were to test shear measurement accuracy across variable point-spread functions from facilities like Subaru Telescope and Hubble Space Telescope, to probe model bias in methods used by teams working on CFHTLenS and KiDS, and to inform error budgets for space missions such as Euclid (spacecraft) and Nancy Grace Roman Space Telescope. Organizers sought to engage researchers from groups at Max Planck Institute for Astrophysics, Caltech, Princeton University, University of Chicago, and industrial partners including teams affiliated with Google and IBM who contributed computational expertise.
The challenge provided multiple simulation branches emulating astronomical imaging pipelines developed by teams at STFC, National Astronomical Observatory of Japan, and Institute of Astronomy, Cambridge. Datasets included variable-shear fields, constant-shear fields, and star fields for point-spread function estimation, reflecting noise regimes and pixel sampling typical of Subaru Telescope, Hubble Space Telescope, and planned Euclid (spacecraft) observations. Participants downloaded image tiles with associated truth catalogs generated using software libraries maintained by groups at University of Oxford, University of Toronto, and Leiden University. The format echoed conventions used in Sloan Digital Sky Survey data releases and cross-matched metadata with coordinate systems employed by Gaia. Submissions were evaluated through leaderboards that compared estimated shear statistics against input shears, with metrics inspired by analyses from CFHTLenS and theoretical frameworks advanced by researchers at Institute for Advanced Study and Princeton University.
Competitors applied a range of techniques drawn from communities centered at Harvard University, Yale University, Columbia University, and University of Michigan. Approaches included model-fitting algorithms such as forward-modeling suites developed by teams at Max Planck Institute for Astrophysics and University of Toronto; moment-based methods refined by groups at University College London and University of Edinburgh; and machine learning pipelines influenced by work at Stanford University, ETH Zurich, and industrial labs like Microsoft Research and DeepMind. Several entries employed Bayesian methods using priors developed in collaboration with researchers at Cambridge University and Imperial College London, while others utilized convolutional neural networks inspired by architectures from MIT and University of Oxford. Star-galaxy separation and PSF interpolation drew on algorithms tested in surveys led by Dark Energy Survey and Pan-STARRS, incorporating calibration techniques from CFHTLenS and theoretical error modeling from Max Planck Institute for Extraterrestrial Physics.
Analysis of submissions revealed trade-offs between bias, variance, and computational cost, mirroring conclusions from teams at University of Bonn and University of Portsmouth. High-performing methods achieved shear recovery consistent with statistical requirements projected for missions like Euclid (spacecraft) and Wide Field Infrared Survey Telescope, while others exposed sensitivity to realistic systematics such as undersampling, detector nonlinearity, and PSF variation encountered in instruments like Hubble Space Telescope and Subaru Telescope. Comparative studies published by collaborators at University College London, ETH Zurich, and Max Planck Institute for Astrophysics quantified multiplicative and additive biases, connecting those to galaxy morphology priors from catalogs like COSMOS and simulation suites produced by teams at INAF. The leaderboard highlighted methodological strengths: model-based inference excelled under high signal-to-noise conditions, whereas machine-learning approaches showed resilience to complex noise distributions but required careful training-set design, an insight echoed by analyses from Caltech and Princeton University.
GREAT10 influenced standards and pipelines adopted by major survey collaborations including Euclid Consortium, Dark Energy Survey, and Kilo-Degree Survey, shaping calibration strategies and informing instrument requirements at agencies such as European Space Agency and NASA. It fostered cross-disciplinary exchanges between astrophysics groups at Max Planck Society, University College London, and machine-learning groups at Stanford University and DeepMind. Subsequent challenges and benchmarking efforts, including successors organized by consortia around LSST Corporation and initiatives affiliated with NASA Ames Research Center, built upon the datasets, metrics, and best-practice guidelines established during GREAT10. The competition accelerated adoption of open-data simulation standards used by teams at University of Oxford and Leiden University and contributed to improved shear pipelines deployed in cosmological analyses by collaborations such as CFHTLenS and KiDS.
Category:Astronomy competitions