arXiv:2610.14533
Sample-Size Scalability and Representation Sensitivity of Quantum Machine Learning Models for Higgs Boson Event Classification
Shiksha Bhardwaj, Ankhi Roy
Abstract
Quantum machine learning offers new approaches to multivariate event classification in high-energy physics, but its performance relative to optimized classical methods remains uncertain under limited training data and restricted quantum resources. We investigate the sample-size scalability and representation sensitivity of a quantum support vector machine and a variational quantum classifier for Higgs-boson signal--background discrimination using simulated events from the ATLAS Higgs Machine Learning Challenge. The models are trained on balanced subsets ranging from 250 to 4000 events and evaluated on held-out samples preserving the original class distribution. Principal component analysis (PCA) and autoencoder (AE) representations are examined to reduce the input dimensionality and quantum-resource requirements. Four optimized classical classifiers are used as benchmarks under a common preprocessing, validation, and evaluation protocol. We assess training-sample scalability in classical and quantum classifiers using a four-component PCA representation and examine representation sensitivity through a secondary matched-event comparison of four- and eight-component PCA representations. Within the evaluated Higgs-boson classification setting, the QSVM showed stronger performance than the VQC at larger training sample sizes, while remaining below the best-performing classical classifiers.
AI Summary
The paper asks how training-set size and dimensionality reduction affect quantum versus classical classifiers for Higgs signal–background discrimination. Using the simulated ATLAS Higgs Challenge dataset, it compares a fidelity-kernel QSVM and a variational quantum classifier (VQC) with logistic regression, RBF–SVM, random forest, and XGBoost. Balanced, nested training sets contain $250$–$4000$ events, with three training seeds. Preprocessing is fitted on training data, configurations and AMS thresholds are selected on validation data, and evaluation uses fixed 800-event test subsets. Quantum calculations use exact, shot-free simulation. With four PCA components, at $N=4000$ the RBF–SVM achieves balanced accuracy $0.7474\pm0.0072$ and event-weighted ROC–AUC $0.8709\pm0.0024$, compared with $0.7303\pm0.0173$ and $0.8529\pm0.0088$ for QSVM. VQC reaches weighted AUC $0.7559\pm0.0299$. QSVM improves more consistently with training size than VQC, without overtaking the leading classical model. XGBoost has the highest mean expanded approximate median significance (AMS), $1.1053$, versus $0.8921$ for QSVM. A paired four-versus-eight-component PCA comparison on a separate test subset finds architecture-dependent effects: at $N=4000$, weighted AUC changes by $+0.0237$ for random forest, $-0.0201$ for QSVM, and $-0.0876$ for VQC. The VQC comparison changes architecture as well as dimension. Stated limitations include only three seeds, descriptive rather than inferential uncertainties, small matched evaluation samples, and no hardware or noise evaluation. The results do not establish quantum advantage, eventual crossover, or mechanisms explaining quantum-model degradation.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.