ML Systems
HGCAL Data Compression & Clustering
Carnegie Mellon University · CMS / CERN · 2024 – 2026
Contributed compression and clustering models for the CMS High-Granularity Calorimeter (HGCAL) — learned autoencoder compression of sensor data at the source and downstream clustering, part of moving intelligence onto the detector edge.
Compressing at the source
The High-Granularity Calorimeter produces far more sensor data than can be read out. The solution is to compress on the detector itself, before transmission — a natural fit for a learned autoencoder that keeps the physically important structure while discarding the rest.
I contributed to compression and clustering models for HGCAL sensor data: an edge-ML problem where the model must be small, quantized, and faithful, because information thrown away here is gone for good.