AI Post Transformers · Visual Companion

Boosted Decision Trees for CMS Muon Triggers

A real-time trigger does not get the luxury of “running ML” the way a datacenter does. This page visualizes the actual co-design move: offline training plus discretization plus a giant lookup table, so online inference collapses into one deterministic memory access inside FPGA timing limits.

Primary Paper PDF Original Viz Link Listen to Podcast

Decision Window

~500 ns inside the Level-1 endcap muon path

Collision Firehose

40 MHz bunch crossings reduced to ~100 kHz

Learned Representation

Offline TMVA BDTs compiled into a ~1.2 GB LUT

Physics Signal

Muon bending: low-pT bends more, high-pT bends less

References

Compact source map used to anchor the visuals. arXiv-style links are included where directly available; otherwise the source links point to the provided paper or scholar query.

1. Boosted Decision Trees for CMS Muon Triggers — PDF

2. Applications and Techniques for Fast Machine Learning in Science (2022) — Scholar

3. Fast inference of Boosted Decision Trees in FPGAs for particle physics (2020) — Scholar

4. Automatic heterogeneous quantization of deep neural networks for low-latency inference on the edge for particle detectors (2021) — Scholar

5. Serving DNNs in Real Time at Datacenter Scale with Project Brainwave (2018) — Scholar

6. The CMS Trigger System — Scholar

7. The CMS Endcap Muon Track Finder — Scholar

8. TMVA: Toolkit for Multivariate Data Analysis — Scholar

9. Fast Machine Learning for Science: how accelerated hardware and software are enabling real-time data analysis at the edge (2022) — Scholar

10. hls4ml: An Open-Source Codesign Workflow to Empower Scientific Low-Power Machine Learning Devices (2021) — Scholar