Joint HEP/Theory Seminar: New machine learning algorithms for the CMS level-1 trigger at the CERN LHC

Recent advancements in computer science and tools are making the ability to deploy machine learning (ML) algorithms in firmware increasingly powerful and accessible. One such tool is hls4ml, a source-to-source compiler for translating ML algorithms into firmware for field-programmable gate arrays (FPGAs). This has exciting implications for the development of novel ML-based level-1 (L1) triggers implemented on FPGAs for particle physics experiments like CMS at the CERN LHC. For example, ML-based anomaly detection methods like variational autoencoders (VAEs) have been gaining popularity as a way of extracting potential new physics signals in a model-agnostic way. These unsupervised algorithms can be trained on unlabeled data rather than simulations, making them a good candidate for anomaly detection triggers. I will describe two VAE-based anomaly detection algorithms, "AXOL1TL" and "CICADA", that are expected to be deployed in the CMS L1 trigger this year as the very first neural-network-based algorithms on CMS L1 FPGAs. We also briefly discuss ideas for future ML trigger algorithms that are expected to be deployed following a similar strategy for the high-luminosity LHC.