AI in Astronomy Workshop

21 May 2025, 09:00 → 23 May 2025, 17:30 Europe/Rome

Aula Magna (INAF-OACT)

The AI in Astronomy workshop is the first in a series of events organized and supported by the USC8-AI Thematic Group of the INAF USC VIII Computing Division.

The primary goal of the AI in Astronomy thematic group is to facilitate the sharing of knowledge, experiences and best practices, provide support, and promote collaboration among INAF researchers who are currently using or plan to use ML/DL techniques in their research. Additionally, the group aims to foster synergies between various astrophysical topics where a specific ML/DL approach could serve as a connecting link.

By pooling their expertise, the group aims to help members identify suitable solutions for specific problems, and tackle the challenges of applying AI models in astronomy - from data preparation and preprocessing to model development, training, testing, optimization, and deployment.

Further information about the group are available here: https://usc8.inaf.it/tematic-groups/ai-in-astronomy/.

• Location: Catania - INAF-Osservatorio Astrofisico
• Date: 21-23 May 2025
• Event Closure: 23 May 2025 at ~16:30
• Registration: No fees; lunches and coffee breaks provided free of charge.
• Participation mode: In-person attendance is strongly recommended (limited to 40-50 participants) in line with the meeting's spirit, easing interactions and opportunities for asking questions. Remote attendance is available for the talk sessions only.
• Workshop Program: The workshop program will feature technical presentations and activity reports from team members, along with training sessions and hands-on activities on selected topics. The talks are meant as “activity reports” from teams or individual researchers, aiming to identify shared interests for potential future projects. Some workshop sessions (discussions, hands-on) will be conducted in Italian. 
  • Day 1: Oral contributions (15-20 minutes maximum) on machine learning activities from team members. Contributions must be submitted in advance through the designated Indico page. Both individual and group contributions are welcome.
  • Day 2-3: Training tutorials consisting of oral presentations and hands-on sessions, delivered by volunteer tutors from the USC8-AI Thematic Group. The tutorials will cover selected machine learning topics, identified through a bottom-up process.  

 

The workshop program is provisional and will be finalized after the abstract submission deadline.

Wednesday 21 May

09:00 → 09:30 Welcome

09:30 → 11:00 Team Activity Reports: Session I

09:30 Napoli Team Activities

A short summary of the activities of the Astroinformatics team from Naples

Speaker: Stefano Cavuoti (INAF - Astronomical Observatory of Capodimonte Napoli)

10:00 Automated identification of diffuse radio emission in all-sky surveys with Radio U-Net

The next generation of radio telescope arrays promises unparalleled sensitivity and resolution, unveiling a wealth of faint and diffuse radio sources in galaxy clusters and beyond. Conventional cataloging approaches struggle with the complexity of such data. To address this, we introduce Radio U-Net, a convolutional neural network based on the U-Net architecture, tailored for identifying diffuse radio sources, such as haloes and relics. Trained on synthetic observations derived from cosmological simulations, Radio U-Net demonstrates robust segmentation capabilities, achieving high accuracy in both detection and morphological recovery of radio sources. In this talk, I will present our network and its application to the LOFAR Two Metre Sky Survey data.

Speaker: Chiara Stuardi (Istituto Nazionale di Astrofisica (INAF))

10:20 Evaluating small vision-language models as AI assistants for radio astronomical source analysis tasks

The advent of next-generation radio telescopes, such as the Square Kilometre Array, promises to revolutionize radio astronomy by generating unprecedented data volumes that challenge traditional processing methods. Deep learning techniques have shown significant potential in tackling various radio analysis tasks, but their effectiveness is often hindered by the scarcity of large, balanced annotated datasets. Recent studies have addressed this limitation through self-supervised learning on unlabelled radio survey data, resulting in foundational radio vision models. These models typically require coding expertise for task adaptation, which limits their broader adoption among astronomers. A text-based interface could overcome this barrier by enabling task-specific queries through examples and customizable outputs.
In this context, Large Language Models (LLMs) have transformed scientific research and daily life with their natural language interfaces and zero-shot learning capabilities. Yet, deploying large-scale models remains resource-intensive and costly. This study investigates small-scale Vision-Language Models (VLMs) as AI assistants for radio astronomy, combining LLM capabilities with vision transformers for image processing. We fine-tuned the LLaVA VLM on a dataset of over 59,000 radio images and instruction queries, evaluating its performance on various radio benchmarks, including source morphology classification, extended source detection, and artifact identification. The resulting model demonstrates clear improvements on radio tasks compared to base models. However, the performance of pure vision models remains unmatched, underscoring the need to improve visual-textual alignment and training dataset quality. This work marks a first step in quantifying the current effectiveness of VLMs, laying a foundation for further developments in radio astronomy.

Speaker: Dr Simone Riggi (INAF - Osservatorio Astrofisico di Catania)

10:40 Machine learning for asterism selection in adaptive optics

Speaker: Fabio Rossi (Istituto Nazionale di Astrofisica (INAF))

11:00 → 11:15 Coffee Break

11:15 → 13:15 Team Activity Reports: Session II

11:15 OATo/UniGe Synergy - Integrating Observations and AI: Transformative Approaches for Unraveling Solar Activity

The Solar Physics group at the Astrophysical Observatory of Torino (OATo), part of the National Institute for Astrophysics (INAF), stands at the forefront of solar physics research. It specializes in investigating the solar wind and eruptive events - from the dynamic layers of the solar atmosphere to the complexities of the inner heliosphere - through a multi-instrument, multi-spacecraft approach. By combining remote-sensing observations with in-situ measurements, it is possible to obtain a comprehensive picture of the physical processes driving Space Weather. In collaboration with the University of Genova (UniGe), which brings extensive expertise in applying Artificial Intelligence (AI) to solar physics and Space Weather forecasting, the OATo Solar Physics research integrates advanced AI methodologies with robust, physics-based models. This synergy ensures that AI-driven predictions are firmly rooted in the fundamental mechanisms governing solar activity. Such a reverse-engineering approach has allowed OATo and UniGe to gain fresh insights into the initiation of solar events, particularly flares and Coronal Mass Ejections (CMEs), areas where many questions still remain unanswered. During this talk, examples of how these integrative strategies - melding state-of-the-art remote-sensing observations with in-situ measurements and incorporating AI-driven approaches - have advanced the understanding of key solar phenomena will be presented. This presentation will showcase how AI has not only improved the interpretation of observational data but has also opened new avenues for exploring largely uncharted aspects of solar activity. The objective is to illustrate the transformative potential of integrating observational expertise with data-driven AI techniques to enhance Space Weather forecasting and to stimulate further exploration in this promising field.

Speaker: Daniele Telloni (Istituto Nazionale di Astrofisica (INAF))

11:35 The MIDA’s activities in space sciences: Artificial intelligence for space weather

This talk provides an overview of the work carried out by the MIDA (Methods for Image and Data Analysis) group at Dipartimento di Matematica, Università di Genova, focusing on the application of artificial intelligence to solar physics and space weather forecasting. Our team combines deep learning techniques with key physical principles to develop models that can predict solar activity and related events. Our research includes the automatic identification and analysis of active regions from solar disk images and magnetograms, the forecasting of solar flare events, the prediction of Coronal Mass Ejection transit times through the integration of remote-sensing coronal images with in-situ solar wind data, and the prediction of geomagnetic storm using solar wind and geomagnetic activity data. With several team members affiliated with the INAF, we see this workshop as an ideal forum to promote collaboration and share new ideas and projects that can advance our mutual efforts not only in space weather research but across various areas of solar and astrophysics and related disciplines.

Speaker: Sabrina Guastavino (Department of Mathematics, Università degli Studi di Genova)

11:55 Extending foreground emission with neural networks

One of the major challenges in the context of the Cosmic Microwave Background (CMB) radiation is to detect a polarization pattern, the so called B-modes of CMB polarization, that are thought to be directly linked to the space-time fluctuations present in the Universe at the very first instants of life. To date, several challenges have prevented to detect the B-modes partly because of the lower sensitivity of the detectors. Our own Galaxy is observed in this context as a foreground contamination. In this talk we will show how novel techniques involving supervised and unsupervised learning encoding can be adopted in order to improve the modeling of the Galactic polarized emission at sub-millimetric wavelengths. This is particularly relevant in order to better characterize the foreground contamination for future CMB experiments (e.g. SO, LiteBIRD, CMB-S4 ), where high sensitivities are expected to be achieved.

Speaker: Giuseppe Puglisi (Universita' di Catania)

12:15 Benchmarking Quantum Convolutional Neural Networks for Signal Classification in Simulated Gamma-Ray Burst Detection

This study evaluates the use of Quantum Convolutional Neural Networks (QCNNs) for identifying signals resembling Gamma-Ray Bursts (GRBs) within simulated astrophysical datasets in the form of light curves. The task addressed here focuses on distinguishing GRB-like signals from background noise in simulated Cherenkov Telescope Array Observatory (CTAO) data, the next-generation astrophysical observatory for very high-energy gamma-ray science. QCNNs, a quantum counterpart of classical Convolutional Neural Networks (CNNs), leverage quantum principles to process and analyze high-dimensional data efficiently. We implemented a hybrid quantum-classical machine learning technique using the Qiskit framework, with the QCNNs trained on a quantum simulator. Several QCNN architectures were tested, employing different encoding methods such as Data Reuploading and Amplitude encoding. Key findings include that QCNNs achieved accuracy comparable to classical CNNs, often surpassing 90\%, while using fewer parameters, potentially leading to more efficient models in terms of computational resources. A benchmark study further examined how hyperparameters like the number of qubits and encoding methods affected performance, with more qubits and advanced encoding methods generally enhancing accuracy but increasing complexity. QCNNs showed robust performance on time-series datasets, successfully detecting GRB signals with high precision. The research is a pioneering effort in applying QCNNs to astrophysics, offering insights into their potential and limitations. This work sets the stage for future investigations to fully realize the advantages of QCNNs in astrophysical data analysis.

Speaker: Farida Farsian (Istituto Nazionale di Astrofisica (INAF))

12:35 Atmospheric Retrievals with Quantum Extreme Learning Machines

The study of exoplanetary atmospheres traditionally relies on forward models to analytically compute the spectrum of an exoplanet by fine-tuning numerous chemical and physical parameters. However, the high-dimensionality of parameter space often results in a significant computational overhead. In this work, we introduce a novel approach to atmospheric retrieval leveraging on quantum extreme learning machines (QELMs). QELMs are quantum machine learning techniques that employ quantum systems as a black box for processing input data. In this work, we propose a framework for extracting exoplanetary atmospheric features using QELMs, employing an intrinsically fault-tolerant strategy suitable for near-term quantum devices, and we demonstrate such fault tolerance with a direct implementation on IBM Fez. The QELM architecture we present shows the potential of quantum computing in the analysis of astrophysical datasets and may, in the near-term future, unlock new computational tools to implement fast, efficient, and more accurate models in the study of exoplanetary atmospheres.

Speaker: Prof. Tiziano Zingales (Università degli Studi di Padova)

13:00 → 14:30 Lunch

14:30 → 15:30 Team Activity Reports: Session III

14:30 Deep Metal: Applying Transformer Models to Photometric Light Curves for Metallicity Estimation

Metallicity is a key parameter in stellar evolution, yet its determination often relies on high-resolution spectra, limiting observational coverage. In this work, we explore the use of Transformer-based models to infer metallicity directly from photometric light curves, leveraging their sequential learning capabilities and attention to temporal patterns. Our results show performance comparable to the best RNN model from our previous work. However, with the upcoming Gaia DR4 data release—providing longer time-series observations—we anticipate significant improvements in accuracy and robustness. This approach paves the way for large-scale metallicity analysis in photometric surveys.

Speaker: Dr Lorenzo Monti (Istituto Nazionale di Astrofisica (INAF))

14:50 Redefining X-ray Polarimetry: Insights from Imaging X- ray Polarimetry Explorer (IXPE) and Future Directions

The Imaging X-ray Polarimetry Explorer (IXPE) has demonstrated that X-ray polarimetry is a powerful tool in astronomy, offering crucial insights into emission mechanisms and the geometry of compact objects. It operates within the 2–8 keV energy range. These measurements are made possible by three polarization-sensitive Gas Pixel Detectors (GPDs) at the core of the focal plane, each equipped with a custom ASIC serving as a charge-collecting anode.However to overcome the limitations we have to take advantage of the recent advances in X-ray optics, gas detectors with different thickness, pressures and gas mixtures would be required. Using next-generation ASICs, like Timepix3, it is possible to have parallel fast readout, providing simultaneous time and charge information for each pixel.For enabling 3D imaging of photo-electron tracks. We explored such a possibility using GridPix detectors.

Speaker: Saba Imtiaz (Istituto di Astrofisica e Planetologia Spaziali, Roma, Italy (INAF))

15:10 Survey of Surveys: Homogenizing Stellar Parameters Across Surveys with Symbolic Regression

Large astronomical surveys play a significant role in modern astrophysics, providing extensive datasets for various studies, i.e. stellar populations and galactic evolution. However, systematic discrepancies between spectroscopic surveys—arising from differences in instrumentation, spectral coverage, and analysis techniques—pose significant challenges for studies that rely on data from multiple catalogs and for reconciling conclusions drawn from different sources. In this context, I am working on a unified catalog of stellar parameters from large spectroscopic surveys such as APOGEE, LAMOST, and GALAH. To identify and correct these biases, I am exploring symbolic regression, a transparent machine learning approach, and comparing it with traditional polynomial fitting. The goal is to provide a more accurate and consistent set of stellar parameters that can be used across different surveys. While this work is ongoing, I will present preliminary results and discuss the challenges of balancing model accuracy with generalization.

Speaker: Aleksandra Avdeeva (Istituto Nazionale di Astrofisica (INAF))

15:30 → 16:00 Coffee Break

16:00 → 18:00 Deep learning theory and application for astrophysics: (I)

Convener: Prof. Simone Palazzo

Thursday 22 May

09:00 → 11:00 Deep learning theory and application for astrophysics: (II)

Convener: Prof. Giovanni Bellitto

11:00 → 11:30 Coffee Break

11:30 → 13:00 Hands-on: Overview on ML Frameworks + PyTorch

Convener: Prof. Matteo Pennisi

13:00 → 14:30 Lunch

14:30 → 16:00 Hands-on: Good practices for ML projects

Convener: Prof. Federica Proietto Salanitri (DIEEI, Università di Catania)

16:00 → 16:30 Coffee Break

16:30 → 17:30 Discussion & Planning

Friday 23 May

09:00 → 10:00 Hands-on: Classification of astronomical images

Convener: Dr Simone Riggi (INAF - Osservatorio Astrofisico di Catania)

10:00 → 11:00 Hands-on: Regression models for astronomical data

Convener: Dr Stefano Cavuoti (INAF - Astronomical Observatory of Capodimonte Napoli)

11:00 → 11:30 Coffee Break

11:30 → 12:30 Hands-on: Time series analysis with astronomical data

Conveners: Dr Lorenzo Monti (Istituto Nazionale di Astrofisica (INAF)), TBD

12:30 → 13:30 Deep learning theory and application for astrophysics: Self-Supervised Learning

Convener: Dr Thomas Cecconello (Istituto Nazionale di Astrofisica (INAF))

13:30 → 15:00 Lunch

15:00 → 16:00 Deep learning theory and application for astrophysics: Graph Neural Networks

Convener: Dr Farida Farsian (Istituto Nazionale di Astrofisica (INAF))

16:00 → 16:30 Coffee Break

16:30 → 17:00 Workshop Closure

Discussion and final remarks