TensoMeta-VQC: A Tensor-Train-Guided Meta-Learning Framework for Robust and Scalable Variational Quantum Computing

Jun Qi, Chao-Han Yang, Pin-Yu Chen, Min-Hsiu Hsieh · 2025

Read TensoMeta-VQC: A Tensor-Train-Guided Meta-Learning Framework for Robust and Scalable Variational Quantum Computing by Jun Qi, Chao-Han Yang, Pin-Yu Chen, Min-Hsiu Hsieh (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-01116.txt.

Introduction

Jun Qi and co-authors hand quantum circuit tuning to a classical tensor network, in this 2025 paper. Variational quantum computing stalls on flat training landscapes and noise. Their TensorHyper-VQC framework lets a tensor-train network generate every circuit parameter, freeing training from the quantum hardware's own gradients.

Theory from learning research backs the design's approximation power and stability. Across dot classification, a classic optimization task, and molecular simulation, the framework beat rivals and shrugged off noise, even in a run on a large IBM quantum processor. The authors present it as a scalable path for near-term quantum learning.

Catalog record

CREATORJun Qi, Chao-Han Yang, Pin-Yu Chen, Min-Hsiu Hsieh
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-01116

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

TensoMeta-VQC: A Tensor-Train-Guided Meta-Learning Framework for Robust and Scalable Variational Quantum Computing Jun Qi, Chao-Han Yang, Pin-Yu Chen, Min-Hsiu Hsieh Abstract Variational Quantum Computing (VQC) faces fundamental barriers in scalability, primarily due to barren plateaus and quantum noise sensitivity. To address these challenges, we introduce TensoMeta-VQC, a novel tensor-train (TT)-guided meta-learning framework designed to improve the robustness and scalability of VQC significantly. Our framework fully delegates the generation of quantum circuit parameters to a classical TT network, effectively decoupling optimization from quantum hardware. This innovative parameterization mitigates gradient vanishing, enhances noise resilience through structured low-rank representations, and facilitates efficient gradient propagation. Based on Neural Tangent Kernel and statistical learning theory, our rigorous theoretical analyses establish strong guarantees on approximation capability, optimization stability, and generalization performance. Extensive empirical results across quantum dot classification, Max-Cut optimization, and molecular quantum simulation tasks demonstrate that TensoMeta-VQC consistently achieves superior performance and robust noise tolerance, establishing it as a principled pathway toward practical and scalable VQC on near-term quantum devices. 1 Introduction Variational Quantum Computing (VQC) [ 1 ] has emerged as a prominent paradigm for exploiting the computational capabilities of Noisy Intermediate-Scale Quantum (NISQ) devices [ 2 , 3 ] . VQC integrates quantum circuits and classical optimization routines to address computationally challenging problems across various domains [ 4 , 5 , 6 , 7 , 8 , 9 ] . Despite its significant promise, the practical realization of VQC’s potential is substantially hindered by fundamental obstacles related to its trainability, robustness against quantum noise, and overall scalability as problem sizes increase [ 10 ] . A primary impediment to scalable VQC is the barren plateau phenomenon [ 11 , 12 , 13 ] , where gradients vanish exponentially with increasing qubit count or circuit depth, significantly degrading result fidelity and destabilizing training [ 14 , 15 , 16 ] . To overcome these critical limitations, as shown in Fig. 1 , we introduce TensoMeta-VQC, a meta-learning [ 17 ] framework predicated on Tensor-Train (TT) decomposition [ 18 ] , engineered to facilitate scalable and robust VQC. The central innovation of TensoMeta-VQC lies in its parameterization strategy. Instead of directly optimizing parameters on the quantum device, a classical TT network [ 19 , 20 ] is trained to generate the parameters for the VQC. Consequently, the entire optimization loop, including gradient computation and parameter updates, is offloaded to the classical TT-cores. In this architecture, the quantum circuit functions solely as a fixed forward-pass evaluator, executing with the parameters supplied by the TT network to provide the objective function value. Figure 1: An illustration of the TensoMeta-VQC framework . A classical TT network generates the parameters (\alpha_{u},\beta_{u},\gamma_{u}) that define the Pauli rotation gates R_{X}(\alpha_{u}),R_{Y}(\beta_{u}),R_{Z}(\gamma_{u}) in a VQC. The TT-generated parameters are injected into the fixed quantum circuit, which operates solely in inference mode. The loss is computed from quantum measurements, and gradients are backpropagated only through the classical TT-cores. Gaussian noise is optionally injected during training to enhance robustness. This architecture decouples optimization from quantum hardware, mitigating barren plateaus and improving noise resilience. This meta-parameterization approach fundamentally reshapes the VQC optimization landscape, offering distinct advantages for scalability and robustness. By confining the optimization process to the classical TT network, TensoMeta-VQC significantly mitigates the barren plateau problem, as gradients are backpropagated through the well-conditioned, deterministic structure of the TT network rather than the potentially deep and complex quantum circuit. The framework’s resilience to quantum noise is also markedly enhanced. Since the trainable parameters reside entirely within the classical domain, gradient updates are insulated from the stochasticity of quantum measurements and hardware imperfections. Moreover, the low-rank structure inherent in the TT representation further contributes to noise suppression by effectively averaging measurement noise across the TT-core gradients, a phenomenon theoretically linked to variance reduction. Our Neural Tangent Kernel (NTK) [ 21 , 22 , 23 , 24 ] analysis supports the improved trainability afforded by this structured parameterization. Comprehensive theoretical analyses and extensive numerical simulations substantiate the efficacy and advantages of TensoMeta-VQC. Our theoretical investigations confirm its enhanced representation capability, trainability, and generalization properties. Experimentally, TensoMeta-VQC demonstrates superior performance and robustness across various tasks, including quantum machine learning applications like quantum dot classification, where it outperforms standard VQC and other classical and hybrid models, even under diverse noise models and circuit configurations. In combinatorial optimization, TensoMeta-VQC-enhanced Quantum Approximate Optimization Algorithms (QAOA) [ 25 , 26 ] achieve superior solution quality for Max-Cut problems compared to conventional QAOA, maintaining stability in noisy simulations. Furthermore, its application to the quantum simulation of the LiH molecule Hamiltonians [ 27 , 28 ] confirms that TensoMeta-VQC achieves improved chemical accuracy and retains strong noise resilience under depolarizing noise, validating its practical applicability. Collectively, these findings establish TensoMeta-VQC as a powerful and versatile computational framework that significantly advances state-of-the-art variational quantum computing. By synergistically integrating concepts from classical tensor network methods and meta-learning principles, TensoMeta-VQC offers a promising path towards realizing more scalable, robust, and ultimately practical quantum algorithms on near-term and future quantum computing platforms. 2 Results 2.1 The TensoMeta-VQC Framework Our proposed framework, TensoMeta-VQC, fundamentally re-engineers the variational quantum computing training paradigm to address significant core scalability and robustness issues. As illustrated in Figure 1 , the core innovation is delegating parameter generation to a classical TT network. In this meta-learning approach, the classical TT network, composed of interconnected low-rank tensor cores, is the sole trainable component. It learns to generate the high-dimensional parameter vectors \hat{\textbf{w}}=[\alpha_{1:U},\beta_{1:U},\gamma_{1:U}]^{\top} required to define the unitary rotation gates of the VQC. The entire operational flow is as follows: 1. Parameter Generation : The classical TT network generates all parameters required for quantum circuits in VQC, effectively shifting parameter optimization entirely to the classical domain. Specifically, given a high-dimensional tensor \mathcal{W}\in\mathbb{R}^{d_{1}\times\cdot\cdot\cdot\times d_{K}} representing quantum circuit parameters, we factorize it into interconnected, low-rank TT-cores as: \mathcal{W}(i_{1},i_{2},...,i_{K})=\mathcal{G}_{1}(i_{1})\mathcal{G}_{2}(i_{2})\cdot\cdot\cdot\mathcal{G}_{K}(i_{K}), (1) with each core \mathcal{G}_{k}(i_{k})\in\mathbb{R}^{r_{k-1}\times r_{k}} characterized by low TT-ranks. Utilizing Gaussian noise input \textbf{z}\sim N(0,I) , the TT network deterministically generates quantum gate parameters \hat{\textbf{w}} as: \hat{\textbf{w}}=\text{TT}(\textbf{z};\{\mathcal{G}_{k}\}_{k=1}^{K}). (2) 2. Quantum Forward Pass : The generated parameters are injected into a fixed quantum circuit, which functions exclusively in inference mode to evaluate the loss. The quantum circuits act as a fixed forward pass evaluator, where no parameters within the VQC are trained. In contrast, gradient computations and parameter updates occur solely in the classical TT-cores via standard backpropagation. This architectural design transforms the optimization landscape, insulating gradient computations from quantum hardware imperfections. It effectively mitigates the barren plateaus barrier and quantum noise-induced instability. 2.2 Theoretical Guarantees We rigorously provide a theoretical foundation for the advantages of the TensoMeta-VQC framework from four perspectives: representation capability, trainability, generalization performance, and robustness against quantum noise. • Approximation Capability . The approximation error \epsilon_{\rm app} , which measures the discrepancy between the expected risk \mathcal{R}(\cdot) of a target function h^{*} and that of the best possible TensoMeta-VQC operator f_{\boldsymbol{\theta}^{*}} , is bounded as: \begin{split}\epsilon_{\rm app}&\leq\mathcal{R}(h^{*})-\mathcal{R}(f_{\boldsymbol{\theta^{*}}})\\ &\leq K_{\ell}\lVert h^{*}(\textbf{x})-f_{\boldsymbol{\theta}^{*}}(\textbf{x})\rVert_{1}\\ &\leq K_{\ell}\lVert h^{*}(\textbf{x})-f_{\textbf{w}^{*}}(\textbf{x})\rVert_{1}+K_{\ell}\lVert f_{\textbf{w}^{*}}(\textbf{x})-f_{\boldsymbol{\theta}^{*}}(\textbf{x})\rVert_{1}\\ &\leq\mathcal{O}\left(K_{\ell}e^{-\alpha L}\right)+\mathcal{O}\left(\frac{K_{\ell}}{2^{\beta U}}\right)+\mathcal{O}(K_{\ell}\epsilon_{\rm tt}),\end{split} (3) where we define the parameters \boldsymbol{\theta}=\{\mathcal{G}_{k}\}_{k=1}^{K} and the best \boldsymbol{\theta}^{*} corresponds to the optimal TT-cores; The loss function \mathcal{R}(\cdot) is assumed as K_{\ell} -Lipschitz continuous and f_{\textbf{w}^{*}}(\textbf{x}) refers to an optimal VQC operator with the best parameters \textbf{w}^{*} .

Read the complete text