Unbinned multivariate observables for global SMEFT analyses from machine learning

A bstract Theoretical interpretations of particle physics data, such as the determination Wilson coefficients Standard Model Effective Field Theory (SMEFT), often involve inference multiple parameters from a global dataset. Optimizing requires identification observables that exhibit highest possible sensitivity to underlying theory parameters. In this work we develop flexible open source frame-work, ML4EFT, enabling integration unbinned multivariate into SMEFT fits. As compared traditional measurements, enhance by preventing information loss incurred when binning in subset final-state kinematic variables. Our strategy combines machine learning regression and classification techniques parameterize high-dimensional likelihood ratios, using Monte Carlo replica method estimate propagate methodological uncertainties. proof concept construct for top-quark pair Higgs+ Z production at LHC, demonstrate their impact on parameter space binned study improved constraints associated inputs. Since number neural networks be trained scales quadratically with can fully parallelized, ML4EFT framework is well-suited which depend up tens EFT coefficients, required