INAF SCHOOL OF AI

14 Apr 2026, 09:00 → 17 Apr 2026, 15:00 Europe/Rome

Auditorium Nazionale Ernesto Capocci (INAF Astronomical Observatory of Capodimonte (Naples))

 

The INAF School of AI is the second in a series of events organized and supported by the AI Thematic Group of the INAF USC-C Computing Division. The primary goal of the thematic group is to facilitate the sharing of knowledge, experiences, and best practices, as well as to provide support and promote collaboration among INAF researchers who are currently using or planning to use machine learning techniques in their research. Further information about the group, including its scope, contacts, and how to join, is available here: https://usc8.inaf.it/tematic-groups/ai-in-astronomy/

 

This four-day school offers an overview of essential machine learning concepts and techniques, structured to build knowledge progressively.

Day 1: Foundations and Data Preparation The school kicks off by establishing a strong foundation. Participants will learn how to approach, visualize, and understand complex datasets using powerful dimensionality reduction techniques (like PCA and t-SNE), the critical steps of data preprocessing, focusing on cleaning, normalization, transformation, and strategies for handling missing data.

Day 2: Supervised Learning We dive into Supervised Learning, exploring how to train models using labeled data. The school covers the core methodology, including train/test/validation splitting and the essential metrics used to evaluate classification and regression models. This is followed by an exploration of classic ML algorithms.

Day 3: Unsupervised Learning The third day focuses on Unsupervised Learning, the science of finding hidden patterns in unlabeled data. Sessions will cover metrics for evaluating clustering quality (e.g., silhouette score) and introduce a range of fundamental clustering algorithms.

Day 4: Anomaly Detection The school concludes with a session on Anomaly Detection. This module will introduce some techniques used to identify outliers and unusual events in data.

The schedule also includes Flash Talk sessions offering participants an opportunity to present their own work and research to their peers.

 

The school can be followed remotely but we strongly suggest to participate in person in order to fully benefit of the presence of the tutors.

With only 100 in-person spots available, we'd greatly appreciate it if you could update your registration should you no longer be able to join us. This helps us ensure everyone who wants to attend has the opportunity to do so.

 

We also inform you that the lessons will be recorded.

Tuesday 14 April

14:00 → 14:30 Welcome and Logistics

14:00 Welcome

Speaker: Pietro Schipani (Istituto Nazionale di Astrofisica (INAF))

14:05 USC-C

Speaker: Andrea Possenti (Istituto Nazionale di Astrofisica (INAF))

14:15 AI Thematic Group

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

14:25 Info & Logistics

Speaker: Giuseppe Riccio (Istituto Nazionale di Astrofisica (INAF))

env_ai_school.yml

14:30 → 15:40 Introduction

Participants will learn how to approach, visualize, and understand complex datasets using powerful dimensionality reduction techniques (like PCA and t-SNE). The afternoon is dedicated to the critical steps of data preprocessing, focusing on cleaning, normalization, transformation, and strategies for handling missing data (imputation).

14:30 Introduction to Machine Learning

• Introduction
• Big data projects in astronomy
• What is ML and how do we use it in astronomy?
• Regression & highlights
• Classification & highlights
• Anomaly detection & highlights
• Object detection & highlights
• Forecasting & highlights
• Data preprocessing & generation & highlights
• Outlier detection
• Simple methods (e.g. IQR-based)

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

1_Riggi-Introduction_ML.pdf

15:00 Data basics & formats

• Data properties (5Vs, format, modality, dimensionality, etc)
• Tabular data formats most used in Astrophysics
• Ascii/csv
• ROOT
• FITS
• HDF5/NetCDF
• VO tables
• Parquet
• Relational DB
• Data I/O considerations

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

2_Farsian-Data_formats.pdf

15:40 → 16:10 Coffee Break

16:10 → 18:00 Introduction

Participants will learn how to approach, visualize, and understand complex datasets using powerful dimensionality reduction techniques (like PCA and t-SNE). The afternoon is dedicated to the critical steps of data preprocessing, focusing on cleaning, normalization, transformation, and strategies for handling missing data (imputation).

16:10 Data Cleaning

• Data consistency checks
• Data visualization
• Examples of 1D data visualization: pie/bar/graph/histogram
• Examples of 2D visualization: scatter plots, 2D histograms (lego, contour, color maps)
• Examples of 3D visualization: volume renderings, iso surface, slicing planes
• Examples of ND data visualization: correlation, scatter plots
• Dimensionality reduction: curse of dimensionality, methods for dimensionality reduction: PCA, t-SNE, UMAP
• Handling missing data
• Missingness mechanisms
• Imputation methods: Listwise/pairwise deletion, mean imputation, regression, MICE/IterativeImputer

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

17:10 Data Transformation

• Tabular data pre-processing
• Linear transforms: minmax normalization, standardization, scaling
• Non-linear transforms: power/log, Box-Cox, Yeo-Johnson, Quantile
• Transforming categorical data

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

17:40 Tutorial - Data Cleaning & Transformation

Speakers: Farida Farsian (Istituto Nazionale di Astrofisica (INAF)), Dr Simone Riggi (INAF - Osservatorio Astrofisico di Catania)

Riggi-Data_Cleaning_Tutorial.html

Riggi-Data_Cleaning_Tutorial.ipynb

Riggi-Data_Cleaning_Tutorial.pdf

Tutorial_Data_Cleaning_Transformation.zip

18:00 → 19:00 Flash Talks

18:00 Quantum Markov Chain Monte Carlo for Cosmological Functions

In this talk, an hybrid quantum algorithm for Bayesian inference will be presented, QMCMC, following the Markow Chain Monte Carlo method. This algorithm proposes the new steps via the evaluation of the statevector of a quantum circuit, then evaluates the objective function and the acceptance rate classically using the Metropolis-Hastings criterium. The QMCMC algorithm has been tested with real cosmological datasets and likelihoods, involving SNe Ia, BAO, and the CMB. The results show contours which are consistently overlapping with those found by classical tools to infer the Bayesian contours for cosmological parameters, proving the reliability of the QMCMC.

Speaker: Giuseppe Sarracino (Istituto Nazionale di Astrofisica (INAF))

Flash_Talk_Naples_Sarracino.pdf

18:10 DeepLandforms: A Deep Learning Computer Vision Toolset Applied to a Prime Use Case for Mapping Planetary Skylights

This work presents DeepLandforms, an open-source toolkit for landform mapping using validated Deep Learning methods. It provides a complete workflow, covering data preparation, model training, and inference, and gives users full control and customization over the entire process. To validate the tool, it was applied to a science case mapping sinkhole-like landforms on Mars, demonstrating its potential for a wide range of future applications.

Speaker: Giacomo Nodjoumi (Istituto Nazionale di Astrofisica (INAF))

Nodjoumi_DeepLandforms_FT_AI_School_2026.pptx

18:20 The GalRSG project: Red Supergiants on the brink of core-collapse

The GalRSG project is a long-term, high-cadence, multi-band photometric monitoring campaign designed to detect pre-supernova variability and intense mass-loss events in Red Supergiants (RSGs), which are key progenitors of Type II supernovae. By applying advanced machine learning techniques to high-precision photometric data, the project aims to identify anomalies and mass-loss signatures in stellar light curves. Given the large volume and multi-band nature of the data, machine learning is essential for efficient analysis. This work will improve our understanding of the late evolutionary stages of massive stars and the physical processes leading up to core-collapse supernova explosions.

Speaker: Mario Lauriano (INAF - Osservatorio Astronomico di Palermo, Università di Padova)

Lauriano_WEB.pdf

Wednesday 15 April

09:30 → 11:00 Supervised Learning

We dive into Supervised Learning, exploring how to train models using labeled data. The session covers the core methodology, including train/test/validation splitting and the essential metrics used to evaluate classification and regression models. This is followed by an exploration of classic ML algorithms.

09:30 Supervised Learning Part I - Theory

theoretical introduction to supervised learning key concepts:

• scaling
• hyper parameter tuning
• train-validation-test split
• cross validation
• classification and regression metrics
• feature importance

the following models will be introduced as well:

• k-nn
• Decision Trees (DT)
• Random Forests (RF)
• Support Vector Machines (SVM)

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

Cavuoti_Supervised_notebook.ipynb

11:00 → 11:30 Coffee Break

11:30 → 13:00 Supervised Learning

We dive into Supervised Learning, exploring how to train models using labeled data. The session covers the core methodology, including train/test/validation splitting and the essential metrics used to evaluate classification and regression models. This is followed by an exploration of classic ML algorithms.

11:30 Supervised Learning Part I - hands-on session

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

Cavuoti_Supervised_notebook.ipynb

Cavuoti_Supervised_notebook.zip

13:00 → 14:15 Lunch & Group Photo

14:15 → 15:30 Supervised Learning

We dive into Supervised Learning, exploring how to train models using labeled data. The session covers the core methodology, including train/test/validation splitting and the essential metrics used to evaluate classification and regression models. This is followed by an exploration of classic ML algorithms.

14:15 Supervised Learning Part II - Theory and Hands-on session

1) Theoretical introduction to Supervised Learning key concepts:

• Perceptron
• Multi Layer Perceptron
• Activation Functions
• Cost Functions
• Optmizers
• Regularization techniques
• Convolutional Neural Network
  • Convolution
  • Pooling
  • building CNNs
  • CNN examples: VGG, ResNet, Inception
• Convolutional Autoencoder
  The lesson includes code examples

During the hands-on session, several exercises will be carried out independently, consisting in the implementation of convolutional neural architectures.

Speaker: Giuseppe Angora (Istituto Nazionale di Astrofisica (INAF))

Angora_INAF_AI_School.pdf

Angora_INAF_AI_School.pptx

SL_WenAfternoon.zip

15:30 → 16:00 Coffee Break

16:00 → 17:30 Supervised Learning

We dive into Supervised Learning, exploring how to train models using labeled data. The session covers the core methodology, including train/test/validation splitting and the essential metrics used to evaluate classification and regression models. This is followed by an exploration of classic ML algorithms.

16:00 Supervised Learning Part II - Hands-on

Please refer to 'Supervised Learning Part II - Theory' in order to download materials as well as to the contribution description

Speaker: Giuseppe Angora (Istituto Nazionale di Astrofisica (INAF))

17:30 → 18:30 Flash Talks

17:30 Machine Learning and Super resolution for radio astronomical data analysis

The main objective of the project is to improve the quality and usability of data from next-generation radio telescopes, such as those involved in the SKA project, by developing advanced methodologies to increase resolution, automatically remove artifacts, and intelligently compress datasets to optimize storage. These datasets, extremely large and multidimensional, require innovative solutions that integrate high-performance computing (HPC), machine learning, and immersive visualization technologies. The expected outcomes include new tools for the analysis and management of astrophysical big data, contributing to scientific progress in astronomy and generating broader technological impacts across various domains.

Speaker: Giorgia Vitanza (INAF - Osservatorio Astrofisico di Catania)

Presentazione Scuola AI Napoli Vitanza.pdf

17:40 The first extragalactic Mg-K anticorrelation

It is well established that Globular Clusters (GCs) are formed by multiple populations (MPs) with different light element abundances (C, N, O, Na, Mg, Al, Si, K), that are structured in well-defined patterns, such as the Na-O and Mg-Al anticorrelations. This evidence is interpreted as the result of a self-enrichment process, where a second population of stars was formed from material processed by massive stars of the first population (polluters) through the hot CNO cycle and its secondary NeNa and MgAl chains. A detailed analysis of these anticorrelations can put constraints on the models for the different proposed polluters. Mg-K anticorrelation is the most extreme manifestation of the phenomenon of MPs in GCs. Such feature was observed only in a handful of Milky Way GCs, and represents a key feature to disentangle among the different polluters proposed in literature. Here I will present the result of the very first Mg-K anticorrelation in an extragalactic GC, namely NGC 1786. Its detection confirms the universality of the MPs and puts further constraints on the theoretical models.

Speaker: Deimer Antonio Alvarez Garay (Istituto Nazionale di Astrofisica (INAF))

flashtalk_Napoli_AI_school_Alvarez.pdf

17:50 Multimodal Foundation Models for Galaxy Classification in Abell S1063

In the era of big-data astronomy, automated and scalable galaxy classification is essential to handle the output of next-generation surveys. This talk presents an application of AstroCLIP, a state-of-the-art cross-modal foundation model, to classify 423 spectroscopically confirmed member galaxies of the cluster Abell S1063 from the CLASH-VLT survey. Leveraging the Transformer architecture and self-supervised contrastive learning, AstroCLIP embeds spectroscopic and photometric data into a shared latent space, allowing the downstream classifier to extract complementary physical information, thus outperforming a single-modal approach. The results highlight the power of foundation models to overcome labeled data scarcity through robust data representations and transfer learning.

Speaker: Lorenzo Santo (Istituto Nazionale di Astrofisica (INAF))

Lorenzo_Santo_INAF_SCHOOL_OF_AI.pptx

18:00 Tracing the Roots of the Early Evolution of the Milky Way through Pulsations (TREE)

Speaker: Emanuela Luongo (Istituto Nazionale di Astrofisica (INAF))

Thursday 16 April

09:30 → 11:00 Unsupervised Learning

The science of finding hidden patterns in unlabeled data. Sessions will cover metrics for evaluating clustering quality (e.g., silhouette score) and introduce a range of fundamental clustering algorithms.

UnsupervisedLearning_partIII.ipynb

UnsupervisedLearning_partIII_exercises.ipynb

09:30 Unsupervised Learning Part I - Theory and Hands-on Session

• Introduction to Unsupervised Learning
• Paradigm and algorithms
• Clustering
  • Overview, workflow, and evaluation metrics
  • K-Means (mechanism, pros and cons, choice of k)
• Hands-on session

Speaker: Ylenia Maruccia (INAF - Astronomical Observatory of Capodimonte)

Unsupervised_partI.ipynb

11:00 → 11:30 Coffee Break

11:30 → 13:00 Unsupervised Learning

The science of finding hidden patterns in unlabeled data. Sessions will cover metrics for evaluating clustering quality (e.g., silhouette score) and introduce a range of fundamental clustering algorithms.

UnsupervisedLearning_partIII.ipynb

UnsupervisedLearning_partIII_exercises.ipynb

11:30 Unsupervised Learning Part II

Self Organizing Maps

Speaker: Francesco Schilliro' (Istituto Nazionale di Astrofisica (INAF))

SOM_01_didattica_base_finale.ipynb

SOM_on_IRIS_finale.ipynb

13:00 → 14:00 Lunch

14:00 → 15:30 Unsupervised Learning

The science of finding hidden patterns in unlabeled data. Sessions will cover metrics for evaluating clustering quality (e.g., silhouette score) and introduce a range of fundamental clustering algorithms.

UnsupervisedLearning_partIII.ipynb

UnsupervisedLearning_partIII_exercises.ipynb

14:00 Unsupervised Learning Part III - Theory

• Clustering algorithms:
  • DBSCAN
  • Hierarchical Clustering
  • HDBSCAN

Speaker: Ylenia Maruccia (INAF - Astronomical Observatory of Capodimonte)

15:30 → 16:00 Coffee Break

16:00 → 17:00 Unsupervised Learning

The science of finding hidden patterns in unlabeled data. Sessions will cover metrics for evaluating clustering quality (e.g., silhouette score) and introduce a range of fundamental clustering algorithms.

UnsupervisedLearning_partIII.ipynb

UnsupervisedLearning_partIII_exercises.ipynb

16:00 Unsupervised Learning Part III - Hands-on Session

This session includes a practical hands-on component.

Speaker: Ylenia Maruccia (INAF - Astronomical Observatory of Capodimonte)

UnsupervisedLearning_partIII.ipynb

UnsupervisedLearning_partIII_exercises.ipynb

18:00 → 19:30 Walking tour of the hystorical center

Friday 17 April

09:30 → 11:00 Anomaly Detection

This module will introduce robust techniques used to identify outliers and unusual events in data.

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

09:30 Anomaly Detection theory

We will analyze four fundamental paradigms for anomaly detection and one model for each of them:

• Distance (k-NN);
• Relative Density (LOF);
• Isolation (Isolation Forest);
• Reconstruction (Autoencoders).

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

1_Cavuoti_Anomaly_Detection.ipynb

11:00 → 11:30 Coffee Break

11:30 → 13:00 Anomaly Detection

This module will introduce robust techniques used to identify outliers and unusual events in data.

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

11:30 Anomaly Detection Hands-on

Five models

• k-NN
• LOF
• Isolation Forest
• One-Class SVM
• Autoencoders

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

1_Cavuoti_Anomaly_Detection.ipynb

13:00 → 14:00 Lunch

14:00 → 15:00 Closing Remarks: Discussion and closing remarks