Publications

What is a Publication?
64 Publications visible to you, out of a total of 64

Abstract (Expand)

Among the many challenges posed by the huge data volumes produced by the new generation of astronomical instruments there is also the search for rare and peculiar objects. Unsupervised outlier detection algorithms may provide a viable solution. In this work we compare the performances of six methods: the Local Outlier Factor, Isolation Forest, k-means clustering, a measure of novelty, and both a normal and a convolutional autoencoder. These methods were applied to data extracted from SDSS stripe 82. After discussing the sensitivity of each method to its own set of hyperparameters, we combine the results from each method to rank the objects and produce a final list of outliers.

Authors: Lars Doorenbos, Stefano Cavuoti, Massimo Brescia, Antonio D'Isanto, Giuseppe Longo

Date Published: 2020

Publication Type: Journal

Abstract (Expand)

This article describes an integration of the subsystems of the distributed parallel simulation environment with cloud infrastructures. A complex support for simulation of the dynamic network object with distributed parameters on Amazon AWS cloud is provided. As well as a tool helping to significantly save running costs for cloud simulations.

Authors: O. Shcherbakov, Kai L. Polsterer, V. A. Svyatnyy

Date Published: 2020

Publication Type: InProceedings

Abstract

Not specified

Authors: F Pozo Nuñez, N Gianniotis, J Blex, T Lisow, R Chini, K L Polsterer, J-U Pott, J Esser, G Pietrzyński

Date Published: 9th Oct 2019

Publication Type: Journal

Abstract

Not specified

Authors: T. J. Galvin, M. Huynh, R. P. Norris, X. R. Wang, E. Hopkins, O. I. Wong, S. Shabala, L. Rudnick, M. J. Alger, K. L. Polsterer

Date Published: 1st Oct 2019

Publication Type: Journal

Abstract

Not specified

Author: Erica Hopkins

Date Published: 1st Oct 2019

Publication Type: Master's Thesis

Abstract (Expand)

The Laplace approximation has been one of the workhorses of Bayesian inference. It often delivers good approximations in practice despite the fact that it does not strictly take into account where the volume of posterior density lies. Variational approaches avoid this issue by explicitly minimising the Kullback-Leibler divergence DKL between a postulated posterior and the true (unnormalised) logarithmic posterior. However, they rely on a closed form DKL in order to update the variational parameters. To address this, stochastic versions of variational inference have been devised that approximate the intractable DKL with a Monte Carlo average. This approximation allows calculating gradients with respect to the variational parameters. However, variational methods often postulate a factorised Gaussian approximating posterior. In doing so, they sacrifice a-posteriori correlations. In this work, we propose a method that combines the Laplace approximation with the variational approach. The advantages are that we maintain: applicability on non-conjugate models, posterior correlations and a reduced number of free variational parameters. Numerical experiments demonstrate improvement over the Laplace approximation and variational inference with factorised Gaussian posteriors.

Author: Nikolaos Gianniotis

Date Published: 1st Jul 2019

Publication Type: InCollection

Abstract (Expand)

The problem of photometric redshift estimation is a major subject in astronomy, since the need of estimating distances for a huge number of sources, as required by the data deluge of the recent years. The ability to estimate redshifts through spectroscopy does not scale with this avalanche of data. Photometric redshifts provide the required redshift estimates at the cost of some precision. The success of several forthcoming missions is highly dependent on the availability of photometric redshifts. The purpose of this thesis is to provide innovative methods for photometric redshift estimation. Two models are proposed. The first is fully-automatized, based on the combination of a convolutional neural network with a mixture density network, to predict probabilistic multimodal redshifts directly from images. The second model is features-based, performing a massive combination of photometric parameters to apply a forward selection in a huge feature space. The proposed models perform very efficiently compared to some of the most common models used in the literature. An important part of the work is dedicated to the correct estimation of the errors and prediction quality. The proposed models are very general and can be applied to different topics in astronomy and beyond.

Author: Antonio D'Isanto

Date Published: 1st Feb 2019

Publication Type: Doctoral Thesis

Powered by
(v.1.15.2)
Copyright © 2008 - 2024 The University of Manchester and HITS gGmbH