Speaker
Description
We present NNPDF4.1, a new release of the NNPDF family that consolidates and extends theoretical, methodological, and experimental progress achieved since NNPDF4.0. On the theory side, predictions for hadron collider observables are now based on state-of-the-art NNLO calculations interfaced to fast interpolation grids, and the analysis systematically incorporates missing higher order uncertainties (MHOUs), approximate N3LO corrections, and QED effects with the photon PDF. On the methodology side, NNPDF4.1 is based on an ensemble of ML models rather than a single one, following a new GPU-accelerated approach to hyperparameter optimisation, and benefit from improved positivity constraints, more general flavour assumptions, and an extended closure and future test validation. The input dataset is enlarged by more than 1200 new data points from 27 new datasets, including several Run II legacy measurements entering a PDF determination for the first time, selected through a systematic dataset selection procedure. We compare NNPDF4.1 to previous NNPDF determinations and to other recent PDF fits, and present a first study of its phenomenological implications.