Ponentes
Descripción
We present a novel deep learning approach, inclusive flavor tagging (IFT), to determine the production flavor of B mesons at the LHCb. This technique is designed to overcome the challenges faced by classical taggers’ performance in the current and future environment, where luminosity and event track multiplicity increase. The IFT utilizes state-of-the-art deep learning models to process information from all particles in the proton-proton collision event, excluding the signal. Our implementation using a DeepSet architecture shows a significant performance gain, increasing the effective tagging power by 35 % for B0 mesons and 20 % for Bs0 mesons over classical taggers, crucial for measurements of flavour oscillation frequencies and time-dependent charge-parity(CP) asymmetries of neutral B mesons. We will also present promising preliminary results from ongoing studies that apply Transformer architectures to this task, highlighting their potential for further improvement.
