The ability of any Machine Learning method to classify the spectra of galaxies depending on the properties of the stellar component rests on the information content of the data. The well-known degeneracies found in population synthesis models suggest this information might be so entangled as to challenge the most sophisticated Deep Learning approaches. This contribution focuses on the traditional definition of entropy to explore this problem from a fundamental viewpoint. We find that the information content – when interpreting the spectrum as a probability distribution function – is reduced to a few spectral intervals that are strongly correlated. Dimensionality reduction via PCA suggests the standard 4000Å break strength and Balmer absorption are the two most informative regions in the analysis of galaxy spectra.

With the volume and availability of astronomical data growing rapidly, astronomers will soon rely on the use of machine learning algorithms in their daily work. This proceeding aims to give an overview of what machine learning is and delve into the many different types of learning algorithms and examine two astronomical use cases. Machine learning has opened a world of possibilities for us astronomers working with large amounts of data, however if not careful, users can trip into common pitfalls. Here we’ll focus on solving problems related to time-series light curve data and optical imaging data mainly from the Deeper, Wider, Faster Program (DWF). Alongside the written examples, online notebooks will be provided to demonstrate these different techniques. This guide aims to help you build a small toolkit of knowledge and tools to take back with you for use on your own future machine learning projects.

Large area astronomical surveys will almost certainly contain new objects of a type that have never been seen before. The detection of ‘unknown unknowns’ by an algorithm is a difficult problem to solve, as unusual things are often easier for a human to spot than a machine. We use the concept of apparent complexity, previously applied to detect multi-component radio sources, to scan the radio continuum Evolutionary Map of the Universe (EMU) Pilot Survey data for complex and interesting objects in a fully automated and blind manner. Here we describe how the complexity is defined and measured, how we applied it to the Pilot Survey data, and how we calibrated the completeness and purity of these interesting objects using a crowd-sourced ‘zoo’. The results are also compared to unexpected and unusual sources already detected in the EMU Pilot Survey, including Odd Radio Circles, that were found by human inspection.

Current and upcoming large optical and near-infrared astronomical surveys have fundamental science as their primary drivers. To cater to those, these missions scan large fractions of the entire sky at multiple wavelengths and epochs. These aspects make these data sets also valuable for investigations into astronomical hazards for life on Earth. The Netherlands Research School for Astronomy (NOVA) is a partner in several optical / near-infrared surveys. In this paper we focus on the astronomical hazard value for two sets of those: the surveys with the OmegaCAM wide-field imager at the VST and with the Euclid Mission. For each of them we provide a brief overview of the astronomical survey hardware, the data and the information systems. We present first results related to the astronomical hazard investigations. We evaluate to what extent the existing functionality of the information systems covers the needs for the astronomical hazard investigations.

Most of our knowledge about the Universe comes from the careful analysis of light that reaches us. Spectroscopy, which is the most detailed method of spectrum analysis, when applied to stars provides information on the parameters of their atmospheres, including effective temperature, acceleration, velocity fields, and their chemical composition. Stellar classification brought forth the understanding of what physical parameters are critical in shaping stellar atmospheres. It is a key element that has linked efforts related to numerical modelling of atmospheres with observations. We present preliminary results on the method of stellar spectra classification based on large-scale unsupervised pre-training. The applied deep neural network of the auto-encoder type, thanks to the use of differentiable elements of physical modelling in the decoder, allows to work with medium to high-resolution spectra, is insensitive to normalization errors, and different radial and rotational velocity, and operates in a wide range of signal-to-noise ratio.

Determination of membership of star clusters is a very important criterion in their study as they effect determination of cluster parameters like radius, age, distance, mass functions, etc. In this paper, we apply the unsupervised method of Gaussian Mixture Modelling (GMM) to find membership of 9 open star clusters of varying ages and locations in the galaxy using Gaia DR3 data. We compare our results to help understand the efficiency of GMM. We find that this method works well with relaxed clusters with ages larger than their relaxation times they approximate Gaussians better.

Machine Learning is a powerful tool for astrophysicists, which has already had significant uptake in the community. But there remain some barriers to entry, relating to proper understanding, the difficulty of interpretability, and the lack of cohesive training. In this discussion session we addressed some of these questions, and suggest how the field may move forward.

During the IAU symposium, 368 “Machine Learning in Astronomy: Possibilities and Pitfalls” in Busan, we organized a panel discussion on the different aspects of data-fusion for large data-sets. Driven by the needs of the scientists, data-fusion technics had been introduces to enable multi-wavelength as well as multi-messenger approaches. This is necessary to get a more detailed and more complete representation of physical phenomena. We identified six different aspects related to data-fusion. Those aspects cover missing data, heterogeneous data, data-access in general, challenges related to data-size, FAIR-data, and future challenges.1

In the field of gravitational-wave (GW) interferometers, the most severe limitation to the detection of transient signals from astrophysical sources comes from transient noise artefacts, known as glitches, that happens at a rate around 1 per minute. Because glitches reduce the amount of scientific data available, there is a need for better modelling and inclusion of glitches in large-scale studies, such as stress testing the search pipelines and increasing the confidence of detection. In this work, we employ a Generative Adversarial Network (GAN) to produce a particular class of glitches (blip) in the time domain. We share the trained network through a user-friendly open-source software package called <monospace>gengli</monospace> and provide practical examples of its usage.

The Pan-STARRS telescopes, located on Haleakala, Maui, Hawaii, are conducting a long-term search of the night sky for Near-Earth Objects. The Pan-STARRS survey is now one of the leading NEO surveys, accounting for more than 40% of all new NEO discoveries, and over 50% of discoveries of larger NEOs. Pan-STARRS is also a prolific comet discovery telescope, and discovered the first interstellar object, ‘Oumuamua.

We proposed a machine learning approach to identify and distinguish dusty stellar sources employing supervised methods and categorizing point sources, mainly evolved stars, using photometric and spectroscopic data collected over the IR sky. Spectroscopic data is typically used to identify specific infrared sources. However, our goal is to determine how well these sources can be identified using multiwavelength data. Consequently, we developed a robust training set of spectra of confirmed sources from the Large and Small Magellanic Clouds derived from SAGE-Spec Spitzer Legacy and SMC-Spec Spitzer Infrared Spectrograph (IRS) spectral catalogs. Subsequently, we applied various learning classifiers to distinguish stellar subcategories comprising young stellar objects (YSOs), C-rich asymptotic giant branch (CAGB), O-rich AGB stars (OAGB), Red supergiant (RSG), and post-AGB stars. We have classified around 700 counts of these sources. It should be highlighted that despite utilizing the limited spectroscopic data we trained, the accuracy and models’ learning curve provided outstanding results for some of the models. Therefore, the Support Vector Classifier (SVC) is the most accurate classifier for this limited dataset.

We present two studies where astrophysics and computational techniques are applied to astroparticle physics and medical image processing. The first one is the use of software as Corsika, Gammapy, Astropy and ARTI to identify potential sites to study astroparticles physics in Ecuador. We determine the secondary particle fluence and M1 signal at different locations, and we calculated the effective area of a prototype Water Cherenkov Detector (WCD) in Quito. The second work is the manipulation of medical images using python and PyDICOM libraries.Aiming to explore the Computer Aided Detection (CAD) systems for the automated detection of numerous lung diseases. We develop a tool for extracting information from DICOM files more efficiently. A second tool consists of a lung segmentation technique for computed tomography images.These works shows the versatility of the computational tools encouraging young researchers to explore and further develop the use of such tools in other areas.

Comets and asteroids collision with Earth and other planets is part of the continued planetary formation, the other part is the solar wind delivers water and gasses back to the Kuiper Belt from the planets, together they form the solar hydrologic cycle. The new theory of solar hydrologic cycle provides that solar wind stripping water and gasses from the inner planets, while having diminished effect on outer planets, is the cause for outer planets becoming gas giants. Jupiter, having the smallest orbit among outer planets, is destined to be the predominant planet and plays a critical role in complex life evolution on Earth. Jupiter grows mass by locking up comet mass, thus reducing the number of comet collisions to Earth. Reduced hydrogen infusion from comets enabled Earth’s atmosphere transitioning from hydrogen to oxygen rich. The transition trajectories of Jupiter mass gain and Earth water and gasses mass fluctuation are calibrated using known geological events. Earth’s trajectory can be divided into three periods, Hydrogen, Carbon Dioxide, and Oxygen, named after predominant gas in Earth’s atmosphere for the period. Complex life only flourishes in the Oxygen Period, when aerobic metabolism is possible. Mass extinction can be caused by cometary hydrogen infusion that incinerates atmospheric oxygen. Probability of such astronomical hazards is declining as outer planets have locked up most comets and will continue to absorb more comets. Earth is safer than ever, and will become even safer, as dictated by the solar water cycle. The physics of the solar hydrologic cycle is universally true, life should be universal phenomena.

Since the last 1990s, Doppler spectroscopy has been one of the most prolific methods of detecting and characterizing exoplanets (Fischer et al. 2016). The latest generation of stabilized spectrographs can achieve impressive levels of precision and stability, approaching that needed to detect the motion of a Sun-like star due to the gravity of an Earth-mass planet in its habitable zone (Crass et al. 2021). However, the exoplanet detection power of modern radial velocity (RV) exoplanet surveys is typically limited by the spectral variability of the target star. Machine learning (ML) has the potential to significantly improve the ability of RV exoplanet surveys to distinguish planets for stellar variability. Astronomers have begun making applying a wide variety of ML techniques, from principal component analysis and multilinear regression to convolutional neural networks. This paper reviews the state of the field for mitigating stellar variability in RV exoplanet surveys from a ML perspective. Early results show that relatively simple ML techniques paired with well-engineered features often perform comparable to much more complex ML models, while providing improved interpretability and explainability. These are likely to be critical factors for establishing the credibility and robustness of any future detections of potentially Earth-like planets.

One of the best and most direct ways to study planet formation processes is to observe young planets while they are forming within their birth protoplanetary disks. As they form, planets tidally interact with their parental disk and produce observable signatures. Recent observations have demonstrated that kinematic planetary signatures (KPS), the perturbed velocity fields of the gas in the protoplanetary disk in the vicinity of the planet, can be observed with the Atacama Large Millimeter/submillimeter Array (ALMA). Here, I introduce a machine learning-based tool KPSFinder (Kinematic Planetary Signature Finder), which aims to find KPS robustly and efficiently.

Migration of bodies under the gravitational influence of almost formed planets was studied, and probabilities of their collisions with the Earth and other terrestrial planets were calculated. Based on the probabilities, several conclusions on the accumulation of the terrestrial planets have been made. The outer layers of the Earth and Venus could accumulate similar planetesimals from different regions of the feeding zone of the terrestrial planets. The probabilities of collisions of bodies during their dynamical lifetimes with the Earth could be up to 0.001-0.01 for some initial semi-major axes between 3.2 and 3.6 AU, whereas such probabilities did not exceed 10−5 at initial semi-major axes between 12 and 40 AU. The total mass of water delivered to the Earth from beyond Jupiter’s orbit could exceed the mass of the Earth’s oceans. The zone of the outer asteroid belt could be one of the sources of the late-heavy bombardment. The bodies that came from the zone of Jupiter and Saturn typically collided with the Earth and the Moon with velocities from 23 to 26 km/s and from 20 to 23 km/s, respectively.