Ponente
Descripción
The LHCb experiment relies on a two-level trigger system to efficiently select events of interest among the vast number of proton-proton collisions that occur at the LHC. In this work, we present a proof-of-concept study exploring the integration of an autoencoder into the High Level Trigger 2 (HLT2) as a novel strategy for event selection. Autoencoders, as unsupervised machine learning algorithms, are capable of learning compact representations of signal events while rejecting background in a model-independent way. This approach offers a key advantage over traditional supervised classifiers such as Boosted Decision Trees, as it does not require explicit background samples, thereby reducing dependence on potentially incomplete or biased background modeling. Using simulated signal data, we train an autoencoder to capture the characteristic features of signal decays, and we demonstrate its ability to identify and reject unseen background-like events. Preliminary results highlight the potential of this method as a tool for signal selection and background suppression, and open the door to further studies on deploying unsupervised machine learning models in real-time selection at LHCb.
Abstract
The LHCb experiment relies on a two-level trigger system to efficiently select events of interest among the vast number of proton-proton collisions that occur at the LHC. In this work, we present a proof-of-concept study exploring the integration of an autoencoder into the High Level Trigger 2 (HLT2) as a novel strategy for event selection. Autoencoders, as unsupervised machine learning algorithms, are capable of learning compact representations of signal events while rejecting background in a model-independent way. This approach offers a key advantage over traditional supervised classifiers such as Boosted Decision Trees, as it does not require explicit background samples, thereby reducing dependence on potentially incomplete or biased background modeling. Using simulated signal data, we train an autoencoder to capture the characteristic features of signal decays, and we demonstrate its ability to identify and reject unseen background-like events. Preliminary results highlight the potential of this method as a tool for signal selection and background suppression, and open the door to further studies on deploying unsupervised machine learning models in real-time selection at LHCb.
