COIN-Focus #1: RESSPECT – France – 2019
The Cosmostatistics Initiative (COIN) is an international network which aims to create an interdisciplinary environment where collaborations between astronomers, statisticians and machine learning experts can flourish. The group utilizes a management model which can find parallel in technological start-ups: based on a dynamic, non-hierarchical and people-centric approach.
Are classification metrics good proxies for SN Ia cosmological constraining power?
We emulate photometric SN Ia cosmology samples with controlled contamination rates of individual contaminant classes and evaluate each of them under a set of classification metrics. We then derive cosmological parameter constraints from all samples under two common analysis approaches and quantify the impact of contamination by each contaminant class on the resulting cosmological parameter estimates. We observe that cosmology metrics are sensitive to both the contamination rate and the class of the contaminating population, whereas the classification metrics are insensitive to the latter. We therefore discourage exclusive reliance on classification-based metrics for cosmological analysis design decisions, e.g. classifier choice, and instead recommend optimizing using a metric of cosmological parameter constraining power.
A graph-based spectral classification of SN-II
This work presents new data-driven classification heuristics for spectral data based on graph theory. As a case in point, we devise a spectral classification scheme of Type II supernova (SNe II) as a function of the phase relative to the V -band maximum light and the end of the plateau phase. Our classification method naturally identifies outliers and arranges the different SNe in terms of their major spectral features. The automated classification naturally reflects the fast evolution of Type II SNe around the maximum light while showcasing their homogeneity close to the end of the plateau phase. The scheme we develop could be more widely applicable to unsupervised time series classification or characterization of other functional data.
Active Learning with RESSPECT
The Recommendation System for Spectroscopic follow-up (RESSPECT) project aims to enable the construction of optimized training samples for the Rubin Observatory Legacy Survey of Space and Time (LSST), taking into account a realistic description of the astronomical data environment. In this work, we test the robustness of active learning techniques in a realistic simulated astronomical data scenario. Our experiment takes into account the evolution of training and pool samples, different costs per object, and two different sources of budget. Results show that traditional active learning strategies significantly outperform random sampling.
Integrated Nested Laplace Approximation (INLA)
We introduce a novel technique to model IFS datasets, which treats the observed galaxy properties as manifestations of an unobserved Gaussian Markov random field. The method is computationally efficient, resilient to the presence of low-signal-to-noise regions, and uses an alternative to Markov Chain Monte Carlo for fast Bayesian inference – the Integrated Nested Laplace Approximation. The proposed Bayesian approach enables the creation of synthetic images, recovery of areas with bad pixels, and an increased power to detect structures in datasets subject to substantial noise and/or sparsity of sampling.