Quantum-Inspired Free-Energy Triggers: Deployable Anomaly Detectors with ML Surrogates and Ultra-Fast Simulation

Real-time analysis is rapidly embracing ML (SMARTHEP whitepaper, 2025), but most model-independent anomaly methods are offline. Building on Matchev et al.’s (2020) Ising test statistic, we propose trigger-level surrogates: compact neural approximators of the Ising free energy computed on partial event views (limited HLT/L1 objects), designed for FPGAs or GPUs. We calibrate and stress-test these surrogates using Lamarr (Barbetti, 2023), which parameterizes detector and reconstruction effects with deep generative models embedded in Gauss, enabling fast and realistic evaluation of trigger rates and false positives. Novelty: (i) a quantum-inspired GOF statistic made deployable at trigger latency; (ii) online-adaptive couplings learned from sliding background estimates to follow run-dependent conditions; (iii) joint optimization for physics robustness and resource constraints using quantized/binarized architectures. By integrating uncertainty-aware thresholds and domain shifts observed in Lamarr, we aim for a trigger that preserves discovery reach for unexpected signals while respecting bandwidth. Impact: expands the search horizon to include anomalies that would otherwise be discarded at trigger time, with quantifiable control of look-elsewhere penalties and computational budgets.

References:

1. A Quantum Algorithm for Model-Independent Searches for New Physics. Konstantin T. Matchev, Prasanth Shyamsundar, Jordan Smolinsky (2020). Letters in High Energy Physics.
2. Review of Machine Learning for Real-Time Analysis at the Large Hadron Collider experiments ALICE, ATLAS, CMS and LHCb. L. Boggia, Carlos Cocha, F. Giasemis, Joachim Hansen, Patin Inkaew, Kaare Endrup Iversen, Pratik Jawahar, Henrique Piñeiro Monteagudo, Micol Olocco, Sten Astrand, M. Borsato, Leon Bozianu, Steven Schramm, the Smarthep Network (2025).
3. Lamarr: LHCb ultra-fast simulation based on machine learning models deployed within Gauss. M. Barbetti (2023). arXiv.org.