A Multiclass Quantum Aligned Centroid Kernel

2026-07-22 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences, Physical Sciences & Chemistry · Depth: Expert, quick

Summary

McQuack is a novel trainable quantum kernel method addressing key limitations of traditional kernel methods. These include quadratic scaling with training set size, fixed kernels, and a lack of intrinsic multiclass formulation. This new approach achieves linear scaling by replacing the full training-set Gram matrix with a trainable sample-to-(class-centroid) fidelity matrix. It was evaluated in simulation and on 124 qubits across two IBM devices using over 150 datasets. In simulation, McQuack demonstrated superior performance against "pure" quantum baselines. Hardware inference, without training, yielded results comparable to an RBF kernel. Trainability studies showed no barren plateaus in experiments up to 13 qubits. This highlights the importance of parameter initialization for successful optimization.

Key takeaway

For Machine Learning Engineers developing quantum kernel methods for multiclass classification, McQuack demonstrates a viable path to overcome quadratic scaling. You should consider implementing trainable sample-to-(class-centroid) fidelity matrices to achieve linear scaling and improve performance. Prioritize careful parameter initialization in your quantum kernel optimization workflows. This is critical for successful model training and avoiding barren plateaus, even on current quantum hardware.

Key insights

McQuack offers a trainable quantum kernel for multiclass classification with linear scaling, outperforming pure quantum baselines.

Principles

• Kernel methods face quadratic scaling and fixed kernels.
• Trainable kernels can achieve linear scaling.
• Parameter initialization is crucial for quantum kernel optimization.

Method

McQuack replaces the full Gram matrix with a trainable sample-to-(class-centroid) fidelity matrix to achieve linear scaling for multiclass quantum classification.

In practice

• Evaluate quantum kernels on IBM devices.
• Use sample-to-centroid fidelity for scaling.
• Prioritize parameter initialization in QML.

Topics

• Quantum Machine Learning
• Kernel Methods
• Multiclass Classification
• Quantum Kernels
• IBM Quantum Devices
• Barren Plateaus

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Benign Overfitting with Quantum Kernels — Local-global quantum kernels enable benign overfitting by combining local and global measurement components.
• An Analytically Trained Variational Surrogate for Quantum Phase Estimation on NISQ Hardware — A classically-trained shallow VQC can accurately surrogate Quantum Phase Estimation on NISQ hardware, bypassing exponential simulation.
• Statevector-Referenced Geometry Survival of a Four-Qubit ZZ Quantum Kernel on IBM Quantum Hardware: A Fixed-Subset Diagnostic Across Three Execution Configurations — Quantum kernel fidelity on hardware does not guarantee task relevance; hardware distortion dominates.
• QCNN with Rough Path Signature Kernels — Hybrid quantum-classical architectures combining QNNs and path signatures can improve time series classification.
• Hash-QNeRF: Multiresolution Hash Encoding for Quantum Neural Radiance Fields — Hybridizing quantum NeRF with multiresolution hash encoding significantly improves convergence and memory efficiency without compromising quantum noise resilience.
• Quantum AI: Quantum Kernel Advantage — Quantum kernels offer a computational advantage over classical AI in high-dimensional pattern recognition, particularly for subtle medical image features.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.