QUIVER: Quantum-Informed Views for Enhanced Representations in Large ML Models
Title: QUIVER: Leveraging Quantum-Informed Perspectives to Boost Representations in Large-Scale Machine Learning Models
Abstract:
The integration of multimodal inputs, which offer complementary perspectives on identical examples, yields significant advantages for large machine learning systems. In this work, we present QUIVER (QUantum-Informed Views for Enhanced Representations), a novel framework that augments traditional data-driven features with a "quantum Fisher view." This view serves as a basis-independent summary of higher-order correlations, derived from a variational quantum circuit (VQC) trained to execute the same classification or regression task.
Distinct from conventional feature augmentation techniques, the quantum Fisher information matrix captures the intrinsic geometric structure of the learned quantum state manifold. Although this feature map—rooted in quantum information theory—is typically challenging to replicate classically, it reveals statistical patterns that are difficult for additional classical data or expanded model capacity to uncover. Consequently, the quantum Fisher view acts as a truly complementary modality rather than a redundant one.
We validate the efficacy of QUIVER by demonstrating performance improvements on standard metrics across two distinct benchmark datasets: QM9, which focuses on molecular property prediction, and JetClass, used for jet flavor classification at the Large Hadron Collider (LHC). The primary contribution of this study is its domain independence; the quantum Fisher view can be integrated into a wide variety of model architectures through specific modifications to the base structure, thereby embedding information regarding the problem’s quantum geometry. These findings indicate that quantum-geometric features, extracted from simulated variational circuits, provide tangible benefits for standard machine learning tasks, achieving measurable value well prior to the development of fault-tolerant quantum hardware.
Source: arXiv Generated at: 2026-06-03 00:00:00 UTC
