|
Caron, S., Gomez-Vargas, G. A., Hendriks, L., & Ruiz de Austri, R. (2018). Analyzing gamma rays of the Galactic Center with deep learning. J. Cosmol. Astropart. Phys., 05(5), 058–24pp.
Abstract: We present the application of convolutional neural networks to a particular problem in gamma ray astronomy. Explicitly, we use this method to investigate the origin of an excess emission of GeV gamma rays in the direction of the Galactic Center, reported by several groups by analyzing Fermi-LAT data. Interpretations of this excess include gamma rays created by the annihilation of dark matter particles and gamma rays originating from a collection of unresolved point sources, such as millisecond pulsars. We train and test convolutional neural networks with simulated Fermi-LAT images based on point and diffuse emission models of the Galactic Center tuned to measured gamma ray data. Our new method allows precise measurements of the contribution and properties of an unresolved population of gamma ray point sources in the interstellar diffuse emission model. The current model predicts the fraction of unresolved point sources with an error of up to 10% and this is expected to decrease with future work.
|
|
|
Panes, B., Eckner, C., Hendriks, L., Caron, S., Dijkstra, K., Johannesson, G., et al. (2021). Identification of point sources in gamma rays using U-shaped convolutional neural networks and a data challenge. Astron. Astrophys., 656, A62–18pp.
Abstract: Context. At GeV energies, the sky is dominated by the interstellar emission from the Galaxy. With limited statistics and spatial resolution, accurately separating point sources is therefore challenging. Aims. Here we present the first application of deep learning based algorithms to automatically detect and classify point sources from gamma-ray data. For concreteness we refer to this approach as AutoSourceID. Methods. To detect point sources, we utilized U-shaped convolutional networks for image segmentation and k-means for source clustering and localization. We also explored the Centroid-Net algorithm, which is designed to find and count objects. Using two algorithms allows for a cross check of the results, while a combination of their results can be used to improve performance. The training data are based on 9.5 years of exposure from The Fermi Large Area Telescope (Fermi-LAT) and we used source properties of active galactic nuclei (AGNs) and pulsars (PSRs) from the fourth Fermi-LAT source catalog in addition to several models of background interstellar emission. The results of the localization algorithm are fed into a classification neural network that is trained to separate the three general source classes (AGNs, PSRs, and FAKE sources). Results. We compared our localization algorithms qualitatively with traditional methods and find them to have similar detection thresholds. We also demonstrate the robustness of our source localization algorithms to modifications in the interstellar emission models, which presents a clear advantage over traditional methods. The classification network is able to discriminate between the three classes with typical accuracy of similar to 70%, as long as balanced data sets are used in classification training. We published online our training data sets and analysis scripts and invite the community to join the data challenge aimed to improve the localization and classification of gamma-ray point sources.
|
|
|
Strege, C., Bertone, G., Besjes, G. J., Caron, S., Ruiz de Austri, R., Strubig, A., et al. (2014). Profile likelihood maps of a 15-dimensional MSSM. J. High Energy Phys., 09(9), 081–59pp.
Abstract: We present statistically convergent profile likelihood maps obtained via global fits of a phenomenological Minimal Supersymmetric Standard Model with 15 free parameters (the MSSM-15), based on over 250M points. We derive constraints on the model parameters from direct detection limits on dark matter, the Planck relic density measurement and data from accelerator searches. We provide a detailed analysis of the rich phenomenology of this model, and determine the SUSY mass spectrum and dark matter properties that are preferred by current experimental constraints. We evaluate the impact of the measurement of the anomalous magnetic moment of the muon (g – 2) on our results, and provide an analysis of scenarios in which the lightest neutralino is a subdominant component of the dark matter. The MSSM-15 parameters are relatively weakly constrained by current data sets, with the exception of the parameters related to dark matter phenomenology (M-1, M-2, mu), which are restricted to the sub-TeV regime, mainly due to the relic density constraint. The mass of the lightest neutralino is found to be < 1.5TeV at 99% C.L., but can extend up to 3 TeV when excluding the g – 2 constraint from the analysis. Low-mass bino-like neutralinos are strongly favoured, with spin-independent scattering cross-sections extending to very small values, similar to 10(-20) pb. ATLAS SUSY null searches strongly impact on this mass range, and thus rule out a region of parameter space that is outside the reach of any current or future direct detection experiment. The best-fit point obtained after inclusion of all data corresponds to a squark mass of 2.3 TeV, a gluino mass of 2.1 TeV and a 130 GeV neutralino with a spin-independent cross-section of 2.4 x 10(-10) pb, which is within the reach of future multi-ton scale direct detection experiments and of the upcoming LHC run at increased centre-of-mass energy.
|
|
|
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.
|
|
|
Stoppa, F., Vreeswijk, P., Bloemen, S., Bhattacharyya, S., Caron, S., Johannesson, G., et al. (2022). AutoSourceID-Light Fast optical source localization via U-Net and Laplacian of Gaussian. Astron. Astrophys., 662, A109–8pp.
Abstract: Aims. With the ever-increasing survey speed of optical wide-field telescopes and the importance of discovering transients when they are still young, rapid and reliable source localization is paramount. We present AutoSourceID-Light (ASID-L), an innovative framework that uses computer vision techniques that can naturally deal with large amounts of data and rapidly localize sources in optical images. Methods. We show that the ASID-L algorithm based on U-shaped networks and enhanced with a Laplacian of Gaussian filter provides outstanding performance in the localization of sources. A U-Net network discerns the sources in the images from many different artifacts and passes the result to a Laplacian of Gaussian filter that then estimates the exact location. Results. Using ASID-L on the optical images of the MeerLICHT telescope demonstrates the great speed and localization power of the method. We compare the results with SExtractor and show that our method outperforms this more widely used method. ASID-L rapidly detects more sources not only in low- and mid-density fields, but particularly in areas with more than 150 sources per square arcminute. The training set and code used in this paper are publicly available.
|
|