Speaker
Description
The LHCb experiment at the LHC employs a fully-software trigger to reconstruct and select events in real time. Key to this approach is the topological beauty (b) trigger, a set of algorithms which select decays of hadrons containing b quarks based on their distinct topology, i.e., highly displaced candidates with a large momentum. For Run 3 of the LHC, these algorithms were reimplemented using Lipschitz monotonic neural network (NN) architectures to provide robustness against varying detector conditions. Selected candidates are used across time-dependent analyses in LHCb, hence the selection must be efficient over and unbiased to the candidate lifetime. This becomes challenging in busy detector environments, wherein several decay processes are present, and the mis-association of constituent particles may result in background candidates with artificially high lifetimes. To this end, distance correlation and moment decomposition approaches to mitigate correlations between candidate lifetime and NN response have been studied and implemented in the LHCb trigger.
This contribution motivates the use of these techniques, compares their respective merits, and discusses their implications for the performance of the LHCb topological b trigger.