The search for rare processes at the Large Hadron Collider has been greatly enhanced by a new framework developed by Marwan Ait Haddou of the University of Casablanca Hassan II, Mohamed Belfkir and Salah Eddin El Harraus of the United Arab Emirates University, and others. Their study introduces a hybrid quantum machine learning (HyQML) approach designed to increase the sensitivity of the search for Higgs boson pairs, an elusive particle central to understanding mass. By combining the power of quantum circuits with classical neural networks, the researchers achieved a two-fold improvement in performance compared to existing methods, allowing for more precise measurements of the Higgs boson’s interactions and tighter constraints on the fundamental parameters governing the universe. This advance could provide new insights into the Higgs boson’s self-association and interactions with other particles, revealing physics beyond the Standard Model.
By improving the discrimination between signal and background, researchers aim to increase the sensitivity of the search for new physics and address critical challenges in particle physics, where identifying rare events requires advanced data analysis techniques. The scientists took advantage of simulated data mirroring conditions at the LHC to represent both the desired Higg spare-generating signal and background processes. They adopted deep neural networks as a baseline and an innovative quantum-classical hybrid model that integrates variational quantum circuits and classical neural networks, leveraging the strengths of both approaches.
Key innovations include meta-learning techniques to address the vanishing gradient problem that hinders quantum machine learning training. The model analyzes high-level kinematic features to distinguish between signal and background. The results show that the hybrid quantum-classical model outperforms classical deep neural networks in separating signal from the background, due to the ability of meta-learning techniques to effectively train quantum circuits. This increased discriminatory power could lead to a potential increase in sensitivity in the search for new physics beyond the Standard Model, allowing scientists to probe deeper into fundamental laws of nature and providing a path to more precise measurements. This study addresses the need for methodological improvements in collider physics, where current exploration leaves room for increased sensitivity to rare particle interactions. The core of this research lies in a hybrid architecture that combines the best of both quantum and classical machine learning.
The researchers used parameterized quantum circuits to map event-level features into a quantum feature space, allowing them to encode data into quantum states and perform transformations that capture complex correlations. At the same time, classical neural networks were integrated to maintain optimization stability and scalability, which is essential for processing large datasets. This combination allows the model to learn complex patterns while remaining computationally manageable. To evaluate the model performance, the HyQML framework was trained to distinguish between signal and background events. The team rigorously compared the HyQML model to state-of-the-art XGBoost models and purely classical implementations and demonstrated a 2x performance improvement.
This leads to tighter constraints on the nonresonant double Higgs boson production cross section and improved estimates of Higgs boson self-coupling and fourth-order vector-particle-Higgs coupling, highlighting the potential of quantum reinforcement learning in collider physics applications. The improved performance will allow scientists to more precisely measure the properties of the Higgs boson and explore deviations from the standard model of particle physics. The research team successfully combined parameterized quantum circuits with classical neural networks to embed event-level features into the feature space while maintaining optimization stability. The results show that the HyQML model outperforms both the state-of-the-art XGBoost model and the pure quantum implementation, with a 2x performance improvement. Analysis of the internal representation of quantum circuits reveals a gradual separation of signal and background distributions during training.
Although early predictions showed overlapping distributions, later epochs showed clear clustering, demonstrating the model’s ability to learn discriminative patterns. To assess statistical sensitivity, the team optimized the scoring area to maximize expected significance. The HyQML model achieves an upper bound on the expected 95% confidence level for the non-resonant double Higgs boson production cross section under background normalized uncertainties and reveals improved constraints on Higgs boson self-coupling and quartic vector-particle-Higgs coupling compared to pure classical and quantum models. The increased sensitivity will allow scientists to more precisely investigate the properties of the Higgs boson and look for subtle deviations from the Standard Model. The research team successfully combined parameterized quantum circuits with classical neural networks to embed event-level features into the feature space while maintaining optimization stability. The HyQML model clearly outperformed both the state-of-the-art XGBoost model and the purely classical implementation, achieving a 2x improvement in performance. This means that the constraints on the nonresonant double Higgs boson production cross section have become stricter.
