2018/06/15 by Dezső Ribli, Ribli, Dezső, Bálint Pataki +3
Computer Science · Mathematics · Physics and Astronomy · #Advanced Image Processing Techniques #Cosmology and Nongalactic Astrophysics (astro-ph.CO) #FOS: Physical sciences #Galaxies: Formation, Evolution, Phenomena #Gaussian Processes and Bayesian Inference #Statistical and numerical algorithms
paper · pdf · doi:10.48550/arxiv.1806.05995
openalex publication_date 2018/06/15 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Dark matter cannot be observed directly, but its weak gravitational lensing\nslightly distorts the apparent shapes of background galaxies, making weak\nlensing one of the most promising probes of cosmology. Several observational\nstudies have measured the effect, and there are currently running, and planned\nefforts to provide even larger, and higher resolution weak lensing maps. Due to\nnonlinearities on small scales, the traditional analysis with two-point\nstatistics does not fully capture all the underlying information. Multiple\ninference methods were proposed to extract more details based on higher order\nstatistics, peak statistics, Minkowski functionals and recently convolutional\nneural networks (CNN). Here we present an improved convolutional neural network\nthat gives significantly better estimates of \Ωm and \σ8\ncosmological parameters from simulated convergence maps than the state of art\nmethods and also is free of systematic bias. We show that the network exploits\ninformation in the gradients around peaks, and with this insight, we construct\na new, easy-to-understand, and robust peak counting algorithm based on the\n'steepness' of peaks, instead of their heights. The proposed scheme is even\nmore accurate than the neural network on high-resolution noiseless maps. With\nshape noise and lower resolution its relative advantage deteriorates, but it\nremains more accurate than peak counting.\n