The ESO-internal ALMA development study, BRAIN1 tackles the ill-posed inverse problem of interferometric astronomical imaging by integrating advanced astrostatistical and astroinformatic methodologies. Beyond improving the imaging process scientifically and operationally, this effort addresses the broader challenge of the data avalanche faced by modern observatories.

The ALMACAL project leverages ALMA maps of calibrator-centered fields to conduct deep mm/sub-mm surveys, enabling the detection of extragalactic sources with flux densities orders of magnitude fainter than achievable with other instruments. These faint sources are critical for refining evolutionary models, as their number counts provide key constraints. In this study, we analyzed band-3 ALMACAL maps from 606 calibrator fields, employing a novel machine learning approach to mitigate the often-overlooked bias introduced by the calibrator itself. Supported by extensive simulations, we extended 100 GHz radio AGN counts by approximately 1.5 orders of magnitude in flux density and refined constraints on dusty star-forming galaxies, reaching sensitivities as low as ∼180 μJy. We have improved the sampling, compared to previous results, in the region of the dominant population transition (between dusty star-forming galaxies and radio AGN). Our results are in good agreement with model predictions.

Deep learning has revolutionized the field of hyperspectral image (HSI) analysis, enabling the extraction of complex spectral and spatial features. While convolutional neural networks (CNNs) have been the backbone of HSI classification, their limitations in capturing global contextual features have led to the exploration of Vision Transformers (ViTs). This paper introduces AMBER, an advanced SegFormer specifically designed for multi-band image segmentation. AMBER enhances the original SegFormer by incorporating three-dimensional convolutions, custom kernel sizes and a Funnelizer layer. This architecture enables to process hyperspectral data directly, without requiring spectral dimensionality reduction during preprocessing. Our experiments, conducted on three benchmark datasets (Salinas, Indian Pines and Pavia University) and on a dataset from the PRISMA ^* satellite, show that AMBER outperforms traditional CNN-based methods in terms of Overall Accuracy, Kappa coefficient, and Average Accuracy on the first three datasets, and achieves state-of-the-art performance on the PRISMA dataset. These findings highlight AMBER’s robustness, adaptability to both airborne and spaceborne data, and its potential as a powerful solution for remote sensing and other domains requiring advanced analysis of high-dimensional data.

The Atacama Large Millimeter/submillimeter Array (ALMA) Calibrators (ALMACAL) project leverages ALMA maps of calibrator-centered fields to conduct deep millimeter/submillimeter surveys, enabling the detection of extragalactic sources with flux densities orders of magnitude fainter than achievable with other instruments. These faint sources are critical for refining evolutionary models, as their number counts provide key constraints. In this study, we analyzed Band 3 ALMACAL maps from 606 calibrator fields, employing a novel machine learning approach to mitigate the often-overlooked bias introduced by the calibrator itself. Supported by extensive simulations, we extended 100 GHz radio active galactic nucleus (AGN) counts by approximately 1.5 orders of magnitude in flux density and refined constraints on dusty star-forming galaxies, reaching sensitivities as low as ∼180 μ Jy. We have improved the sampling, compared to previous results, in the region of the dominant population transition (between dusty star-forming galaxies and radio AGN). Our results are in good agreement with model predictions.

Various approaches have been proposed to describe the geomorphology of drainage networks and the intricate relationships between abiotic/biotic factors and their surrounding environment. There is an intrinsic complexity of the explicit qualification of the morphological variations in response to various types of control factors and the difficulty of expressing the cause-effect links. Traditional methods of drainage network classification are based on the manual extraction of key characteristics, subsequently applied to pattern recognition frameworks. These attitudes, however, have low predictive and uniform ability. For this reason, we present a different approach, based on the data-driven supervised learning by images, extended also to extraterrestrial cases. Using deep learning models, the extraction and classification phases are integrated within a more objective, analytical, and automatic toolkit (Donadio et al., 2021). Pre - processing of satellite and topographical images through image segmentation methodsExtraterrestrial and terrestrial drainage pattern analysis is a topic of central interest since it allows scientists to understand the hydrogeological and geomorphological past of planets and satellites. Earth, Mars, Venus, and Titan’s patterns have been taken into consideration in this work. The extraction process, necessary to obtain an outline of the river, can be done through image segmentation, considering different mathematical methods whose efficiency varies according to the conditions and properties of the images. The segmentation methods used can reliably identify objects’ contours if in the foreground, separating them from the background. In the context of image processing, this is addressed as Edge Detection.Two types of images were addressed, respectively, topographic and satellite. In both, an extensive pre-processing phase has been carried out, to reduce background noise with computationally efficient and optimized algorithms. This makes the profiles of the drainage networks stand out from the rest of the image, minimizing the loss of important information and the need for human intervention.This aims to make the preparatory phase of the images (pre-processing) as self-consistent as possible, to be effectively applied to large volumes of images, allowing the generation of a valid training set for the classification of drainage patterns using self-adaptive methods, based on machine and deep learning paradigms.In the final work, a good trade-off has been achieved between efficiency and effectiveness of the edge-detection methods. As will be discussed later, the need for an expert’s intervention is extremely limited, in most cases not needed at all. River Zoo survey and classification based on Deep Learning modelsThis work introduces an innovative approach to river hydrographic basins classification within the River Zoo Survey project. The main goal is to perform a statistical evaluation of the classification of terrestrial and extraterrestrial drainage networks by human experts to be subsequently used as base of knowledge to train supervised Artificial Intelligence (AI) methods.The idea is to analyze the degree of reliability of class assignment to drainage samples, driven domain expert decisions, based on visual inspection of images and the identification of the right pattern type. Through the analysis of Earth, Mars and Titan’s rivers, experts were asked to classify rivers into one of ten distinct patterns, further categorized into two macro-classes: dendritic and non-dendritic.Figure 1 – Different classes of drainage patterns: a) dendritic; b) sub-dendritic; c) pinnate; d) parallel; e) radial; f) rectangular; g) trellis; h) angular; i) annular; j) contorted. (a)–(c) patterns are related to dendritic forms (D), (d)–(j) to non-dendritic ones (ND).The purpose of this study is to establish an objective classification system for rivers, improving the understanding of terrestrial and extra-terrestrial rivers drainage networks. Using statistical techniques, the study explores methods to reduce noise in human based classification, thus providing a robust classification system for the automatic processing of river data with a detail much better (10 classes) than the current two class classification systems (dendritic and non-dendritic).This work focuses on the methodology and objectives of the research, highlighting its interdisciplinary nature and potential contributions to a better comprehension of river morphologies across different planetary bodies. Classification of drainage patterns using properties of fractalsMachine Learning (ML) models often require a differentiated and big enough training sequence so that they can output a good prediction rule, a predictor, to then use in labeling unseen elements belonging to the testing set. While experts can give their opinions on a given river in relation to the ten classes here considered, it is more of a subjective truth than a ground truth. To minimize the model’s bias, the training sequence should contain data as accurately labeled as possible, thus having a great probability of minimizing the true error.Introducing fractal geometry, involving self-similar objects with a fixed degree of complexity, often referred to as Hausdorff Dimension (HD). By computing the HD for a given set of rivers, they can be grouped into classes by defining step thresholds determining the belonging to any of the ten categories. The classification can be further refined by considering the Horton-Strahler number, which is a way of establishing a hierarchy between tributaries in a drainage network. This makes it possible to keep track of the branching of rivers, along with their intrinsic fractal complexity.To compute the HDs, different methods will be used, leveraging the flexibility and the capabilities of the Python programming language. The best algorithm to compute the HD on rivers, Box Counting (BC), will be analyzed, and compared to the experts’ classification, to further comprehend the thought process of a human mind when presented with a classification task involving complex and branched structures.Results show that fractal analysis is reliable in the context of geomorphology and river patterns, allowing for the creation of a ground truth for the RiverZoo images, and laying the basis for the development of advanced ML algorithms used for classification purposes.

The Euclid mission is expected to image millions of galaxies at high resolution, providing an extensive dataset with which to study galaxy evolution. Because galaxy morphology is both a fundamental parameter and one that is hard to determine for large samples, we investigate the application of deep learning in predicting the detailed morphologies of galaxies in Euclid using Zoobot, a convolutional neural network pretrained with 450 000 galaxies from the Galaxy Zoo project. We adapted Zoobot for use with emulated Euclid images generated based on Hubble Space Telescope COSMOS images and with labels provided by volunteers in the Galaxy Zoo: Hubble project. We experimented with different numbers of galaxies and various magnitude cuts during the training process. We demonstrate that the trained Zoobot model successfully measures detailed galaxy morphology in emulated Euclid images. It effectively predicts whether a galaxy has features and identifies and characterises various features, such as spiral arms, clumps, bars, discs, and central bulges. When compared to volunteer classifications, Zoobot achieves mean vote fraction deviations of less than 12% and an accuracy of above 91% for the confident volunteer classifications across most morphology types. However, the performance varies depending on the specific morphological class. For the global classes, such as disc or smooth galaxies, the mean deviations are less than 10%, with only 1000 training galaxies necessary to reach this performance. On the other hand, for more detailed structures and complex tasks, such as detecting and counting spiral arms or clumps, the deviations are slightly higher, of namely around 12% with 60 000 galaxies used for training. In order to enhance the performance on complex morphologies, we anticipate that a larger pool of labelled galaxies is needed, which could be obtained using crowd sourcing. We estimate that, with our model, the detailed morphology of approximately 800 million galaxies of the Euclid Wide Survey could be reliably measured and that approximately 230 million of these galaxies would display features. Finally, our findings imply that the model can be effectively adapted to new morphological labels. We demonstrate this adaptability by applying Zoobot to peculiar galaxies. In summary, our trained Zoobot CNN can readily predict morphological catalogues for Euclid images.

We present Deep Focus (DF) results on the ALMA image deconvolution problem for point-like sources. We compare DF reconstruction capabilities and execution times with those of CLEAN and other Deep Learning algorithms, showing improvements in reconstruction capabilities with respect to both, and an average speed-up factor of 280 with respect to CLEAN.

We performed differential number counts down to 4.25 sigma using ALMA Band 3 calibrator images, which are known for their high dynamic range and susceptibility to various types of contamination. Estimating the fraction of contaminants is an intricate process due to correlated non-Gaussian noise, and it is often compounded by the presence of false positives generated during the cleaning phase. In addition, calibrator extensions further complicate the counting of background sources. In order to address these challenges, our strategy employs a machine learning-based approach utilizing the UMLAUT algorithm. UMLAUT assigns a value to each detection, and it considers how likely it is for there to be a genuine background source or a contaminant. With respect to this goal, we provide UMLAUT with eight observational input parameters, each automatically weighted using a gradient descent method. Our methodology significantly improves the precision of differential number counts, thus surpassing conventional techniques, including visual inspection. This study contributes to a better understanding of radio sources, particularly in the challenging sub-5 sigma regime, within the complex context of a high dynamic range of ALMA calibrator images.

The Atacama Large Millimeter/submillimeter Array with the planned electronic upgrades will deliver an unprecedented amount of deep and high resolution observations. Wider fields of view are possible with the consequential cost of image reconstruction. Alternatives to commonly used applications in image processing have to be sought and tested. Advanced image reconstruction methods are critical to meet the data requirements needed for operational purposes. Astrostatistics and astroinformatics techniques are employed. Evidence is given that these interdisciplinary fields of study applied to synthesis imaging meet the Big Data challenges and have the potentials to enable new scientific discoveries in radio astronomy and astrophysics.

This work is focused on a deep learning model–U-Net convolutional neural network–with the purpose of segmenting relevant imagery classes, for detecting mining areas using hyperspectral images of the PRISMA Earth Observation mission, funded by the Italian Space Agency (ASI). To avoid the typical problem of hyperspectral data redundancy and to improve the computational performances without losing accuracy, the Singular Value Decomposition (SVD) is applied to the hyperspectral data cube, taking only the first three singular values, thus projecting the multi-dimensional data cube to a three channels image. The method is applied to a PRISMA surface reflectance scene of South-West Sardinia, one of the oldest mining districts in the world. The Quadrilátero Ferrífero mining district (Minas Gerais, Brazil) will also be analyzed to test the transferability of the model to other mining areas worldwide.

An ESO internal ALMA development study, BRAIN, is addressing the ill-posed inverse problem of synthesis image analysis employing astrostatistics and astroinformatics. These emerging fields of research offer interdisciplinary approaches at the intersection of observational astronomy, statistics, algorithm development, and data science. In this study, we provide evidence of the benefits of employing these approaches to ALMA imaging for operational and scientific purposes. We show the potential of two techniques, RESOLVE and DeepFocus, applied to ALMA calibrated science data. Significant advantages are provided with the prospect to improve the quality and completeness of the data products stored in the science archive and overall processing time for operations. Both approaches evidence the logical pathway to address the incoming revolution in data rates dictated by the planned electronic upgrades. Moreover, we bring to the community additional products through a new package, ALMASim, to promote advancements in these fields, providing a refined ALMA simulator usable by a large community for training and/or testing new algorithms.

ABSTRACT We present a deep learning (DL) pipeline developed for the detection and characterization of astronomical sources within simulated Atacama Large Millimeter/submillimeter Array (ALMA) data cubes. The pipeline is composed of six DL models: a convolutional autoencoder for source detection within the spatial domain of the integrated data cubes, a Recurrent Neural Network (RNN) for denoising and peak detection within the frequency domain, and four residual neural networks (ResNets) for source characterization. The combination of spatial and frequency information improves completeness while decreasing spurious signal detection. To train and test the pipeline, we developed a simulation algorithm able to generate realistic ALMA observations, i.e. both sky model and dirty cubes. The algorithm simulates always a central source surrounded by fainter ones scattered within the cube. Some sources were spatially superimposed in order to test the pipeline deblending capabilities. The detection performances of the pipeline were compared to those of other methods and significant improvements in performances were achieved. Source morphologies are detected with subpixel accuracies obtaining mean residual errors of 10−3 pixel (0.1 mas) and 10−1 mJy beam−1 on positions and flux estimations, respectively. Projection angles and flux densities are also recovered within 10 per cent of the true values for 80 and 73 per cent of all sources in the test set, respectively. While our pipeline is fine-tuned for ALMA data, the technique is applicable to other interferometric observatories, as SKA, LOFAR, VLBI, and VLTI.

The Square Kilometre Array Observatory (SKAO) will explore the radio sky to new depths in order to conduct transformational science. SKAO data products made available to astronomers will be correspondingly large and complex, requiring the application of advanced analysis techniques to extract key science findings. To this end, SKAO is conducting a series of Science Data Challenges, each designed to familiarise the scientific community with SKAO data and to drive the development of new analysis techniques. We present the results from Science Data Challenge 2 (SDC2), which invited participants to find and characterise 233245 neutral hydrogen (Hi) sources in a simulated data product representing a 2000~h SKA MID spectral line observation from redshifts 0.25 to 0.5. Through the generous support of eight international supercomputing facilities, participants were able to undertake the Challenge using dedicated computational resources. Alongside the main challenge, `reproducibility awards' were made in recognition of those pipelines which demonstrated Open Science best practice. The Challenge saw over 100 participants develop a range of new and existing techniques, with results that highlight the strengths of multidisciplinary and collaborative effort. The winning strategy -- which combined predictions from two independent machine learning techniques to yield a 20 percent improvement in overall performance -- underscores one of the main Challenge outcomes: that of method complementarity. It is likely that the combination of methods in a so-called ensemble approach will be key to exploiting very large astronomical datasets.

We present a Deep Learning pipeline for the detection of astronomical sources within radiointerferometric simulated data cubes. Our pipeline is constituted by two Deep Learning models: a Convolutional Autoencoder for the detection of sources within the spatial domain of the cube, and a RNN for the denoising and detection of emission peaks in the frequency domain. The combination of spatial and frequency information allows for higher completeness and helps to remove false positives. The pipeline has been tested on simulated ALMA observations achieving better performances and faster execution times with respect to traditional methods. The pipeline can detect 92% of sources up to a flux of 1.31 Jy/beam with no false positives thus providing a reliable source detection solution for future astronomical radio surveys.

The light curve of Gaia23bab (=SPICY 97589) shows two significant (Δ G > 2 mag) brightening events, one in 2017 and an ongoing event starting in 2022. The source’s quiescent spectral energy distribution indicates an embedded ( A V > 5 mag) pre-main-sequence star, with optical accretion emission and mid-infrared disk emission. This characterization is supported by the source’s membership in an embedded cluster in the star-forming cloud DOBASHI 1604 at a distance of 900 ± 45 pc. Thus, the brightening events are probable accretion outbursts, likely of EX Lup-type.

Traditional supervised classification models aim to approximate the functional mapping between instance attributes and their class labels. These models, however, do not consider the interdependence between instances and global characteristics of data and thus often they lead to poor classification results. In this work, we present a novel hybrid classification model – named HyCASTLE – designed to solve the main shortcomings of hybrid models that employ topological information through clustering in order to improve classifiers performances: they make hypotheses on the underlying data distribution and do not consider the effect of noise. HyCASTLE utilises a non-parametric estimator to capture the underlying data distribution and creates entirely data-driven shape-free clusters. HyCASTLE then refines this cluster configuration using both data topology and available labels through an iterative cluster aggregation and separation process. We evaluated HyCASTLE performance on 37 datasets and compare it with both traditional and hybrid classification models. Our results show that HyCASTLE has comparable or better performance than the other models and results to be more resilient to class noise.

We present a citation pattern analysis between astronomical papers and 13 other disciplines, based on the arXiv database over the past decade (2010–2020). We analyze 12,600 astronomical papers citing over 14,531 unique publications outside astronomy. Two striking patterns are unraveled. First, general relativity recently became the most cited field by astronomers, a trend highly correlated with the discovery of gravitational waves. Second, the fast growth of referenced papers in computer science and statistics, the first with a notable 15 fold increase since 2015. Such findings confirm the critical role of interdisciplinary efforts involving astronomy, statistics, and computer science in recent astronomical research.

The objective of this study was to assess the effectiveness and safety of dupilumab in treating elderly patients with atopic dermatitis from baseline to 52 weeks. A retrospective observational real-life study was conducted in a group of elderly patients with severe atopic dermatitis treated with dupilumab for 52 weeks. Inclusion criteria were: age ≥ 65 years; diagnosis of atopic dermatitis made by an expert dermatologist; Eczema Area and Severity Index ≥ 24; and a contraindication, side effects, or failure to respond to cyclosporine. The primary outcome was the mean percentage reduction in the Eczema Area and Severity Index score from baseline to week 52. Secondary measures included the mean percentage reduction in the Pruritus and Sleep Numerical Rating Scales and the Dermatology Life Quality Index, and the types and rates of adverse events from baseline to week 52. One hundred and five patients were eligible for the study. Flexural dermatitis was the most frequent clinical phenotype (63.8%). The coexistence of more than one clinical phenotype was found in 70/105 (66.6%) patients. We observed a reduction in all disease severity scores from baseline to week 52 (p < 0.001). Adverse events were recorded in 30/105 (28.6%) patients, with conjunctivitis and injection-site reaction the most frequent. In this study, dupilumab is an effective and safe treatment for the long-term management of atopic dermatitis in patients aged over 65 years.

Astrometric detection involves a precise measurement of stellar positions, and is widely regarded as the leading concept presently ready to find earth-mass planets in temperate orbits around nearby sun-like stars. The TOLIMAN space telescope[39] is a low-cost, agile mission concept dedicated to narrow-angle astrometric monitoring of bright binary stars. In particular the mission will be optimised to search for habitable-zone planets around Alpha Centauri AB. If the separation between these two stars can be monitored with sufficient precision, tiny perturbations due to the gravitational tug from an unseen planet can be witnessed and, given the configuration of the optical system, the scale of the shifts in the image plane are about one millionth of a pixel. Image registration at this level of precision has never been demonstrated (to our knowledge) in any setting within science. In this paper we demonstrate that a Deep Convolutional Auto-Encoder is able to retrieve such a signal from simplified simulations of the TOLIMAN data and we present the full experimental pipeline to recreate out experiments from the simulations to the signal analysis. In future works, all the more realistic sources of noise and systematic effects present in the real-world system will be injected into the simulations.

Several approaches were proposed to describe the geomorphology of drainage networks and the abiotic/biotic factors determining their morphology. There is an intrinsic complexity of the explicit qualification of the morphological variations in response to various types of control factors and the difficulty of expressing the cause-effect links. Traditional methods of drainage network classification are based on the manual extraction of key characteristics, then applied as pattern recognition schemes. These approaches, however, have low predictive and uniform ability. We present a different approach, based on the data-driven supervised learning by images, extended also to extraterrestrial cases. With deep learning models, the extraction and classification phase is integrated within a more objective, analytical, and automatic framework. Despite the initial difficulties, due to the small number of training images available, and the similarity between the different shapes of the drainage samples, we obtained successful results, concluding that deep learning is a valid way for data exploration in geomorphology and related fields.