QISK: Quantum-Inspired Streaming Kernels for Robust Classification under Concept Drift

Keywords: Quantum‑Inspired Kernels, Streaming Classification, Concept Drift, Distribution Shift Robustness

TL;DR: QISK is a classical, quantum‑inspired, product‑state kernel with lightweight importance weighting that boosts accuracy under concept drift

Abstract: Streaming binary classifiers suffer performance degradation under concept drift when data distributions change over time. We propose QISK (Quantum-Inspired Streaming Kernels), a quantum-inspired approach that integrates advanced drift detection, quantum kernel ensembles, and enhanced importance weighting for improved worst-case performance under distribution shift. Our method combines multiple quantum-inspired kernels with different parameterizations, advanced ensemble drift detection techniques, and multi-method density ratio estimation, implemented entirely through classical computation. The key innovations include an ensemble of quantum-inspired kernels, advanced DRO-Lite with multiple density ratio estimators, and sophisticated drift detection mechanisms. Experimental evaluation demonstrates improvements in worst-case performance, with QISK achieving 12-14\% absolute improvements over state-of-the-art baselines.

Submission Number: 65