Speaker
Description
Koexai is a deep-tech company working at the interface between machine learning and astrophysical research, supporting scientific teams in translating domain-specific problems into robust ML workflows. Over the last year, within the INAF / PNRR / ICSC Spoke 3 context, Koexai has worked on five research projects spanning supernova characterisation, cosmological simulations, pulsar-timing inference, X-ray background mitigation, and gamma-ray transient analysis.
This contribution uses ASTRAI as a case study of the Koexai approach to Astro-ML research. Developed in collaboration with the University of Catania, ASTRAI starts from physics-based models of hydrogen-rich supernovae and uses their synthetic light curves to train machine-learning models for light-curve generation and physical-parameter inference. The resulting pipeline enables fast characterisation while preserving physical modelling as the foundation of the ML system.
Drawing on ASTRAI and lessons from the other four projects, the talk will discuss how astrophysical knowledge and machine-learning expertise can be combined across problem formulation, data preparation, modelling, validation, and scientific interpretation, highlighting a shared-ownership model for collaboration between research groups and specialised ML teams.