Quantum Kernel Methods for Seasonal Modelling

Workshop Description

Full-day workshop on quantum kernel methods for logistics seasonal modelling. Covers quantum feature maps, QSVM for seasonal pattern classification, regime change detection, and SKU clustering with honest comparisons against classical kernel baselines.

Seasonal patterns in logistics demand are rarely simple sine waves. Weekly cycles interact with monthly billing periods, annual holidays shift dates, and market disruptions create structural breaks that invalidate historical seasonality. Classical seasonal decomposition (STL, X-13) and Fourier methods assume regularity that logistics data often lacks. Quantum kernel methods offer an alternative approach for non-linear pattern detection. Quantum feature maps (Havlicek et al. 2019) encode classical data into exponentially large Hilbert spaces where inner products capture correlations invisible to classical kernels. This workshop examines where this theoretical advantage translates to practical benefit for logistics seasonal modelling. Participants build QSVM classifiers for seasonal pattern detection, regime change identification, and SKU clustering by seasonal profile. Each model is benchmarked against classical RBF-SVM, random forest, and XGBoost on matched datasets. The workshop is direct about limitations: current NISQ hardware restricts feature maps to roughly 10-20 input dimensions, kernel matrix computation scales as O(n^2), and for most logistics seasonal patterns, classical kernels provide equivalent separation. The value is in understanding precisely where quantum kernels might offer advantage as hardware improves.

What participants cover

• Quantum kernel theory: feature maps, quantum kernel estimation, and the advantage hypothesis (Havlicek et al. 2019, Liu et al. 2021)
• Seasonal pattern classification: using QSVM to identify non-linear seasonality in logistics demand with multi-scale temporal features
• Regime change detection: quantum kernels for structural breaks in seasonal behaviour after market disruptions or supply chain reconfiguration
• SKU clustering by seasonal profile: grouping products with similar seasonal behaviour in high-dimensional feature spaces
• Classical benchmarking: rigorous comparison against RBF-SVM, polynomial kernels, random forests, and XGBoost on matched test sets
• NISQ constraints: feature dimension limits, O(n^2) kernel matrix cost, and when classical kernels provide equivalent accuracy