Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards a Self-Driving Trigger at the LHC: Adaptive Response in Real Time

Shaghayegh Emami, Cecilia Tosciri, Giovanna Salvi, Zixin Ding, Yuxin Chen, Abhijith Gandrakota, Christian Herwig, David W. Miller, Jennifer Ngadiuba, Nhan Tran

TL;DR

The paper tackles the challenge of real-time, high-rate data filtering at the LHC by proposing a self-driving trigger framework that autonomously adapts thresholds and reallocates bandwidth in hardware-aware feedback loops. It combines a traditional HT-based trigger with a neural network–based anomaly detector and validates the approach on CMS Open Data, using both simulated and real collision data. The authors develop a simple PD controller for single-path rate stabilization and a multi-path, cost-aware framework with three case studies to balance background, signal efficiency, bandwidth, and compute costs. Across simulations and actual CMS data, adaptive trigger control stabilizes rates, preserves or boosts signal yield, and reduces computational burden, highlighting a practical route toward autonomous triggers in high-throughput experiments.

Abstract

Real-time data filtering and selection -- or trigger -- systems at high-throughput scientific facilities such as the experiments at the Large Hadron Collider (LHC) must process extremely high-rate data streams under stringent bandwidth, latency, and storage constraints. Yet these systems are typically designed as static, hand-tuned menus of selection criteria grounded in prior knowledge and simulation. In this work, we further explore the concept of a self-driving trigger, an autonomous data-filtering framework that reallocates resources and adjusts thresholds dynamically in real-time to optimize signal efficiency, rate stability, and computational cost as instrumentation and environmental conditions evolve. We introduce a benchmark ecosystem to emulate realistic collider scenarios and demonstrate real-time optimization of a menu including canonical energy sum triggers as well as modern anomaly-detection algorithms that target non-standard event topologies using machine learning. Using simulated data streams and publicly available collision data from the Compact Muon Solenoid (CMS) experiment, we demonstrate the capability to dynamically and automatically optimize trigger performance under specific cost objectives without manual retuning. Our adaptive strategy shifts trigger design from static menus with heuristic tuning to intelligent, automated, data-driven control, unlocking greater flexibility and discovery potential in future high-energy physics analyses.

Towards a Self-Driving Trigger at the LHC: Adaptive Response in Real Time

TL;DR

The paper tackles the challenge of real-time, high-rate data filtering at the LHC by proposing a self-driving trigger framework that autonomously adapts thresholds and reallocates bandwidth in hardware-aware feedback loops. It combines a traditional HT-based trigger with a neural network–based anomaly detector and validates the approach on CMS Open Data, using both simulated and real collision data. The authors develop a simple PD controller for single-path rate stabilization and a multi-path, cost-aware framework with three case studies to balance background, signal efficiency, bandwidth, and compute costs. Across simulations and actual CMS data, adaptive trigger control stabilizes rates, preserves or boosts signal yield, and reduces computational burden, highlighting a practical route toward autonomous triggers in high-throughput experiments.

Abstract

Real-time data filtering and selection -- or trigger -- systems at high-throughput scientific facilities such as the experiments at the Large Hadron Collider (LHC) must process extremely high-rate data streams under stringent bandwidth, latency, and storage constraints. Yet these systems are typically designed as static, hand-tuned menus of selection criteria grounded in prior knowledge and simulation. In this work, we further explore the concept of a self-driving trigger, an autonomous data-filtering framework that reallocates resources and adjusts thresholds dynamically in real-time to optimize signal efficiency, rate stability, and computational cost as instrumentation and environmental conditions evolve. We introduce a benchmark ecosystem to emulate realistic collider scenarios and demonstrate real-time optimization of a menu including canonical energy sum triggers as well as modern anomaly-detection algorithms that target non-standard event topologies using machine learning. Using simulated data streams and publicly available collision data from the Compact Muon Solenoid (CMS) experiment, we demonstrate the capability to dynamically and automatically optimize trigger performance under specific cost objectives without manual retuning. Our adaptive strategy shifts trigger design from static menus with heuristic tuning to intelligent, automated, data-driven control, unlocking greater flexibility and discovery potential in future high-energy physics analyses.
Paper Structure (18 sections, 7 equations, 33 figures)

This paper contains 18 sections, 7 equations, 33 figures.

Table of Contents

  1. Introduction
  2. Datasets and Simulation Framework
  3. Features of the Trigger Data and Algorithms
  4. Anomaly Detection Algorithm
  5. Performance of a Fixed Trigger Menu in Simulated Events
  6. Real-time Control for Individual Trigger Rules
  7. Multi Trigger Control Framework
  8. Case 1: Background-Signal Tradeoff
  9. Case 2: Specific Path Bandwidth Allocation
  10. Case 3: Computational Cost Awareness
  11. A Simple Local Controller
  12. Controller Performance with Data
  13. Single-Path Real-Time Control on Data
  14. Multi-Path Real-Time Control on Data
  15. Summary Plots and Data--Simulation Comparison
  16. ...and 3 more sections

Figures (33)

  • Figure 1: Schematic overview of the adaptive trigger control system. A cost-aware agent receives batches of collision data, evaluates trigger performance and computational costs, and updates L1 trigger thresholds and bandwidth allocations in real time through a feedback-driven control loop.
  • Figure 2: The average number of primary vertices ($N_{\mathrm{PV}}$) reconstructed at various points throughout Run 283408 collected by the CMS Experiment in 2016. Time is measured as the fraction of the run that has elapsed, with the gradual decrease of $N_{\mathrm{PV}}$ reflecting the gradual drop in luminosity and pileup as the fill progresses.
  • Figure 3: Evolution of $H_\mathrm{T}$ during a 2016 LHC run recorded by CMS (Run 283408). (a) Histograms show the progressive shift of the $H_\mathrm{T}$ distribution toward lower values over the course of the fill. (b) Violin plots display the evolution of the median and quartiles of $H_\mathrm{T}$ as a function of the run-time fraction, highlighting the correlation between $H_T$, time, and $N_{\mathrm{PV}}$ through variations in the pileup energy density.
  • Figure 4: Introducing a time index in simulated events by ordering them according to the $N_{\mathrm{PV}}$ distribution, smeared with Poisson noise to mimic realistic fluctuations. (a) $N_{\mathrm{PV}}$ versus time, illustrating how the constructed index reproduces the characteristic pileup evolution seen during an LHC fill. (b) Violin plots of $H_\mathrm{T}$ versus time, showing the expected decrease of hadronic activity over the fill as the pileup drops.
  • Figure 5: Comparison of key per-event features across different samples: real collision data, minimum bias simulation, SM $t\bar{t}$ events, and an exotic benchmark process $h\to 4b$. These distributions provide insight into the kinematic and topological differences that the trigger strategies aim to capture.
  • ...and 28 more figures