Differentiable Quantum Architecture Search in Quantum-Enhanced Neural Network Parameter Generation

TL;DR

This work tackles the challenge of deploying quantum-enhanced neural networks by removing reliance on quantum hardware during inference and compressing parameters via Quantum-Train (QT). It introduces Differentiable Quantum Architecture Search (DiffQAS) to automatically discover QNN architectures that generate classical neural network parameters, optimizing architecture and weights end-to-end. Across classification, time-series prediction, and reinforcement learning, DiffQAS-QT matches or exceeds manually designed baselines while massively reducing parameter counts and improving training stability. The results suggest a scalable pathway for practical hybrid quantum-classical AI systems with broad applicability and reduced hardware dependency.

Abstract

The rapid advancements in quantum computing (QC) and machine learning (ML) have led to the emergence of quantum machine learning (QML), which integrates the strengths of both fields. Among QML approaches, variational quantum circuits (VQCs), also known as quantum neural networks (QNNs), have shown promise both empirically and theoretically. However, their broader adoption is hindered by reliance on quantum hardware during inference. Hardware imperfections and limited access to quantum devices pose practical challenges. To address this, the Quantum-Train (QT) framework leverages the exponential scaling of quantum amplitudes to generate classical neural network parameters, enabling inference without quantum hardware and achieving significant parameter compression. Yet, designing effective quantum circuit architectures for such quantum-enhanced neural programmers remains non-trivial and often requires expertise in quantum information science. In this paper, we propose an automated solution using differentiable optimization. Our method jointly optimizes both conventional circuit parameters and architectural parameters in an end-to-end manner via automatic differentiation. We evaluate the proposed framework on classification, time-series prediction, and reinforcement learning tasks. Simulation results show that our method matches or outperforms manually designed QNN architectures. This work offers a scalable and automated pathway for designing QNNs that can generate classical neural network parameters across diverse applications.

Figures (14)

• Figure 1: Comparison between standard VQC and VQC used in the Quantum-Train (QT) framework. The upper circuit illustrates a conventional variational quantum circuit, which consists of an encoding block $U(\vec{x})$ for input-dependent state preparation, followed by a variational block composed of layered parametrized unitaries $\{ V_\ell(\vec{\theta}_\ell)\}$. In contrast, the lower diagram shows the VQC design employed in QT, where the encoding circuit is omitted. By initializing directly from the ground state $\ket{0}^{\otimes n}$, the QT-based VQC reduces circuit depth and resource requirements, facilitating more efficient quantum execution.
• Figure 2: Concept of Quantum-Train.
• Figure 3: Ansatzes of QNN/VQC considered in this work.
• Figure 4: Quantum-Train with DiffQAS
• Figure 5: Comparison of Various QNN Architectures on Binary Classification of Digits 1 and 5 from the MNIST Dataset.