Building a common analysis framework with artificial antelligence


    Data obtained by different labs are usually difficult to merge and used in a single analysis framework. Different experimental conditions as well as kinematic domains covered by experiments and different data format prevent an easy way to combine different data sets (even if referring to the same final state).
    We are exploring the possibility of using Artificial Intelligence to define a common analysis framework. A neural network can be trained using data collected in different experiments and covering different parts of the phase space to provide an equivalent of the cross section in an extended kinematic domain. In principle the same idea can be applied to different reactions to perform a full couple channel analysis, extending the application domain of the network.
    We started developing this idea by analyzing a specific channel that has been measured in different labs (e.g. Jefferson Lab and Mainz). As a template we chose the reaction gamma p \to p pi+ pi- since it is complicated enough (a five differential phase space) and rich in contents (hadron resonance in both baryon and meson system) but still simple enough to be extensively studied. Data form 6 GeV g11 CLAS experiment were selected and used to train a Features Augmentation anPreliminaryt results show that the procedure converges and event generated by the FAT-GAN reproduced the one dimensional projections fo the original data set. This is only the first step toward a full analysis framework since to compare data form different experiments an unfolding procedure is needed (the original data set is affected by the specific response of the detector).
    This is a novel way to extract the five-differential cross section in form of a neural network that could be then trained with another data set to extend the kinematic domain. As further step in the analysis we expect to fit the network to amplitudes derived within Task 1 and Task 2 in order to extract the relevant physics parameters form the data.
    Regular update on the status of the analysis framework completion will be reported in this page.

    This project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 824093.

    Please publish modules in offcanvas position.