NSBI for Proton Structure

NSBI for Proton Structure: We present a new approach to determine the proton structure at the LHC using Neural Simulation Based Inference (NSBI), moving beyond traditional analyses based on binned observables.

Instead of compressing measurements into histograms (and inevitably losing information), we directly exploit the full, high-dimensional structure of collider events. By leveraging AI-driven inference, this allows us to extract PDFs with improved precision and reduced information loss.

As a proof of concept, we apply this framework to top-quark pair production and demonstrate clear gains over standard (binned, low dimensional) methodologies. In turn, this analysis benefits other gluon-initiated channels, with direct impact on LHC flagship measurements such as Higgs production via gluon fusion, where PDF uncertainties remain a key limiting factor.

More broadly, our work contributes toward an ongoing shift in collider physics: from binned dimensional observables to fully unbinned, high-dimensional, ML-assisted measurements, unlocking more of the information already present in the data.

Many thanks for my wonderful collaborators, a real Dream Team of theorists, experimentalists, and ML experts which has worked really hard in the last months to realise a idea which arose during the discussions with Robert Schoefbeck at the last NNPDF Collaboration meeting in Morimondo 2025.

https://inspirehep.net/literature/3145457