Stoppa, F., Bhattacharyya, S., Ruiz de Austri, R., Vreeswijk, P., Caron, S., Zaharijas, G., et al. (2023). AutoSourceID-Classifier Star-galaxy classification using a convolutional neural network with spatial information. Astron. Astrophys., 680, A109–16pp.
Abstract: Aims. Traditional star-galaxy classification techniques often rely on feature estimation from catalogs, a process susceptible to introducing inaccuracies, thereby potentially jeopardizing the classification's reliability. Certain galaxies, especially those not manifesting as extended sources, can be misclassified when their shape parameters and flux solely drive the inference. We aim to create a robust and accurate classification network for identifying stars and galaxies directly from astronomical images.Methods. The AutoSourceID-Classifier (ASID-C) algorithm developed for this work uses 32x32 pixel single filter band source cutouts generated by the previously developed AutoSourceID-Light (ASID-L) code. By leveraging convolutional neural networks (CNN) and additional information about the source position within the full-field image, ASID-C aims to accurately classify all stars and galaxies within a survey. Subsequently, we employed a modified Platt scaling calibration for the output of the CNN, ensuring that the derived probabilities were effectively calibrated, delivering precise and reliable results.Results. We show that ASID-C, trained on MeerLICHT telescope images and using the Dark Energy Camera Legacy Survey (DECaLS) morphological classification, is a robust classifier and outperforms similar codes such as SourceExtractor. To facilitate a rigorous comparison, we also trained an eXtreme Gradient Boosting (XGBoost) model on tabular features extracted by SourceExtractor. While this XGBoost model approaches ASID-C in performance metrics, it does not offer the computational efficiency and reduced error propagation inherent in ASID-C's direct image-based classification approach. ASID-C excels in low signal-to-noise ratio and crowded scenarios, potentially aiding in transient host identification and advancing deep-sky astronomy.
|
|
Stoppa, F., Ruiz de Austri, R., Vreeswijk, P., Bhattacharyya, S., Caron, S., Bloemen, S., et al. (2023). AutoSourceID-FeatureExtractor Optical image analysis using a two-step mean variance estimation network for feature estimation and uncertainty characterisation. Astron. Astrophys., 680, A108–14pp.
Abstract: Aims. In astronomy, machine learning has been successful in various tasks such as source localisation, classification, anomaly detection, and segmentation. However, feature regression remains an area with room for improvement. We aim to design a network that can accurately estimate sources' features and their uncertainties from single-band image cutouts, given the approximated locations of the sources provided by the previously developed code AutoSourceID-Light (ASID-L) or other external catalogues. This work serves as a proof of concept, showing the potential of machine learning in estimating astronomical features when trained on meticulously crafted synthetic images and subsequently applied to real astronomical data.Methods. The algorithm presented here, AutoSourceID-FeatureExtractor (ASID-FE), uses single-band cutouts of 32x32 pixels around the localised sources to estimate flux, sub-pixel centre coordinates, and their uncertainties. ASID-FE employs a two-step mean variance estimation (TS-MVE) approach to first estimate the features and then their uncertainties without the need for additional information, for example the point spread function (PSF). For this proof of concept, we generated a synthetic dataset comprising only point sources directly derived from real images, ensuring a controlled yet authentic testing environment.Results. We show that ASID-FE, trained on synthetic images derived from the MeerLICHT telescope, can predict more accurate features with respect to similar codes such as SourceExtractor and that the two-step method can estimate well-calibrated uncertainties that are better behaved compared to similar methods that use deep ensembles of simple MVE networks. Finally, we evaluate the model on real images from the MeerLICHT telescope and the Zwicky Transient Facility (ZTF) to test its transfer learning abilities.
|
|
de los Rios, M., Petac, M., Zaldivar, B., Bonaventura, N. R., Calore, F., & Iocco, F. (2023). Determining the dark matter distribution in simulated galaxies with deep learning. Mon. Not. Roy. Astron. Soc., 525(4), 6015–6035.
Abstract: We present a novel method of inferring the dark matter (DM) content and spatial distribution within galaxies, using convolutional neural networks (CNNs) trained within state-of-the-art hydrodynamical simulations (Illustris-TNG100). Within the controlled environment of the simulation, the framework we have developed is capable of inferring the DM mass distribution within galaxies of mass similar to 10(11)-10(13)M(circle dot) from the gravitationally baryon-dominated internal regions to the DM-rich, baryon-depleted outskirts of the galaxies, with a mean absolute error always below approximate to 0.25 when using photometrical and spectroscopic information. With respect to traditional methods, the one presented here also possesses the advantages of not relying on a pre-assigned shape for the DM distribution, to be applicable to galaxies not necessarily in isolation, and to perform very well even in the absence of spectroscopic observations.
|
|
Gammaldi, V., Zaldivar, B., Sanchez-Conde, M. A., & Coronado-Blazquez, J. (2023). A search for dark matter among Fermi-LAT unidentified sources with systematic features in machine learning. Mon. Not. Roy. Astron. Soc., 520(1), 1348–1361.
Abstract: Around one-third of the point-like sources in the Fermi-LAT catalogues remain as unidentified sources (unIDs) today. Indeed, these unIDs lack a clear, univocal association with a known astrophysical source. If dark matter (DM) is composed of weakly interacting massive particles (WIMPs), there is the exciting possibility that some of these unIDs may actually be DM sources, emitting gamma-rays from WIMPs annihilation. We propose a new approach to solve the standard, machine learning (ML) binary classification problem of disentangling prospective DM sources (simulated data) from astrophysical sources (observed data) among the unIDs of the 4FGL Fermi-LAT catalogue. We artificially build two systematic features for the DM data which are originally inherent to observed data: the detection significance and the uncertainty on the spectral curvature. We do it by sampling from the observed population of unIDs, assuming that the DM distributions would, if any, follow the latter. We consider different ML models: Logistic Regression, Neural Network (NN), Naive Bayes, and Gaussian Process, out of which the best, in terms of classification accuracy, is the NN, achieving around 93 . 3 per cent +/- 0 . 7 per cent performance. Other ML evaluation parameters, such as the True Ne gativ e and True Positive rates, are discussed in our work. Applying the NN to the unIDs sample, we find that the de generac y between some astrophysical and DM sources can be partially solved within this methodology. None the less, we conclude that there are no DM source candidates among the pool of 4FGL Fermi-LAT unIDs.
|
|
HAWC Collaboration(Alfaro, R. et al), & Salesa Greus, F. (2022). Validation of standardized data formats and tools for ground-level particle-based gamma-ray observatories. Astron. Astrophys., 667, A36–12pp.
Abstract: Context. Ground-based gamma-ray astronomy is still a rather young field of research, with strong historical connections to particle physics. This is why most observations are conducted by experiments with proprietary data and analysis software, as is usual in the particle physics field. However, in recent years, this paradigm has been slowly shifting toward the development and use of open-source data formats and tools, driven by upcoming observatories such as the Cherenkov Telescope Array (CTA). In this context, a community-driven, shared data format (the gamma-astro-data-format, or GADF) and analysis tools such as Gammapy and ctools have been developed. So far, these efforts have been led by the Imaging Atmospheric Cherenkov Telescope community, leaving out other types of ground-based gamma-ray instruments. Aims. We aim to show that the data from ground particle arrays, such as the High-Altitude Water Cherenkov (HAWC) observatory, are also compatible with the GADF and can thus be fully analyzed using the related tools, in this case, Gammapy. Methods. We reproduced several published HAWC results using Gammapy and data products compliant with GADF standard. We also illustrate the capabilities of the shared format and tools by producing a joint fit of the Crab spectrum including data from six different gamma-ray experiments. Results. We find excellent agreement with the reference results, a powerful confirmation of both the published results and the tools involved. Conclusions. The data from particle detector arrays such as the HAWC observatory can be adapted to the GADF and thus analyzed with Gammapy. A common data format and shared analysis tools allow multi-instrument joint analysis and effective data sharing. To emphasize this, a sample of Crab nebula event lists is made public with this paper. Because of the complementary nature of pointing and wide-field instruments, this synergy will be distinctly beneficial for the joint scientific exploitation of future observatories such as the Southern Wide-field Gamma-ray Observatory and CTA.
|
|