Modern astronomical datasets continue to grow exponentially in the era of large sky surveys such as LOFAR, Gaia, EHT, ALMA, Euclid, SDSS-V, ZTF, MeerKAT, SKA, DSA-2000, ELT, Rubin observatory, and Roman mission. The size and complexity of datasets from these and similar astronomical facilities have necessitated adoption of automated approaches for data extraction, analysis, classification, and for helping astronomers draw inferences from data.

These automated approaches including supervised and unsupervised machine learning, and artificial intelligence are transforming our understanding of the universe. AI's ability to process vast and multidimensional datasets efficiently, detect subtle patterns, and automate complex analyses is revolutionizing how we analyze data and study celestial phenomena — from exoplanet discovery to cosmological simulations.

In this respect, foundation models — large neural networks trained on specific tasks using a huge amount of data — are transforming the way we do science, as they have transformed language and internet searches. The secret behind this transformation is their excellent ability to generalize. An emerging question is: can we build a single, large-scale foundation model for astronomy that can excel at many relevant tasks and accelerate scientific discovery? We will discuss the computational cost of such methods, model opacity and avenues for more sustainable advancement of the field, as we continue to profit from the ever-growing data repositories that are becoming available.

While these models offer remarkable new capabilities, many machine learning techniques are still treated as black boxes. We must therefore ask how a model arrives at its conclusions, and whether those conclusions are driven by real astrophysical signals or by hidden biases in the data. Understanding the answer is only half the challenge — we also need to quantify how uncertain that answer is. Probabilistic methods such as Bayesian neural networks, deep ensembles, and simulation-based inference help us characterize what the models do not know and reveal how uncertainties propagate into the final scientific results. Together, interpretability and uncertainty quantification ensure that AI-driven discoveries remain physically meaningful, reliable, and scientifically trustworthy. A part of this symposium will also be dedicated to these use cases and community discussions on effective use, limitations and guidelines for ethical use of AI in astronomy research.

We invite submissions to this session that focus on artificial intelligence applications, machine learning techniques, statistical methods, and AI-centric computational practices.

Programme

We envision themed blocks such as:

• Large scale foundation models and multi-modality
• Self supervised big transformation based NNs
• AI assistants
• Reliable inference (probabilistic ML, uncertainty quantification) and simulation based inference
• Interpretability (including mechanistic interpretability)
• Data at scale (platforms, tools, infrastructure and storage)
• AI applications and use cases in astronomy

Invited speakers

TBD

Scientific organisers

• Aleksandra Avdeeva (INAF - Osservatorio Astrofisico di Arcetri, Italy; Co-chair)
• Amruta Jaodand (Smithsonian Astrophysical Observatory, Centre for Astrophysics, Harvard, USA; Co-chair)
• Daniel Schaerer (Astronomy Dept, University of Geneva; Co-chair)
• John Wenskovitch (Virginia Tech, USA)
• David Cornu (LUX - Observatoire Paris, France)
• Cecilia Garraffo (Centre for Astrophysics, Harvard, USA)
• Rafael Martinez-Galarza (Centre for Astrophysics, Harvard, USA)
• Anna Scaife (University of Manchester, UK)
• Elisabeth Sola (Institute of Astronomy, Cambridge, UK)
• Guillaume Thomas (Instituto de Astrofísica de Canarias, Spain)
• Alessio Turchi (INAF - Osservatorio Astrofisico di Arcetri, Italy)
• Guillaume Guiglion (Zentrum für Astronomie der Universität Heidelberg, Germany)

Contact

aiforastro.eas2026 @ gmail.com