Improving AXOL1TL Trigger Anomaly Detection in CMS Level-1 Trigger

In the search for the optimal triggers to avoid overlooking potential new physics information in proton collision events at the LHC, which are constrained by a low-level DAQ latency of 50 nanoseconds at CMS experiment, unsupervised machine learning (ML) has gained prominence as an efficient tool for encoding the Standard Model background and identifying anomalies using out-of-distribution metrics. AXOL1TL is a trigger located within the µGT system of the CMS Level-1 Trigger currently deployed. We demonstrate that the inclusion of tau leptons into the model, along with adjustments to the dense model architecture, improves sensitivity to rare SM signals or potential BSM escenarios, as well as anomaly classification compared to the baseline.