Princeton University Donald R. Hamilton Colloquium Series

Abstract: As the LHC moves into its high-luminosity era, discovery potential will hinge not just on data volume, but on the intelligence of real-time selection systems. The CMS experiment processes petabytes of data every second, using an FPGA-based trigger system to identify rare signatures amid overwhelming backgrounds. Recent advances incorporate unsupervised machine learning to flag anomalous events within nanoseconds. This talk highlights recent developments in trigger-level anomaly detection—including FPGA-deployed autoencoders, graph networks, and quantized transformers—and outlines how these technologies enable scalable, model-driven and model-independent searches at the HL-LHC and future colliders.