Quantum Surrogate-Driven Image Classifier: A Gradient-Free Approach to Avoid Barren Plateaus

TL;DR

This paper introduces a gradient-free quantum neural network training method using surrogate models and mid-circuit measurements, effectively avoiding barren plateaus and improving image classification accuracy.

Contribution

The paper presents a novel surrogate-driven framework with mid-circuit measurement and reset, mitigating barren plateaus in quantum neural networks for the first time.

Findings

Achieves higher accuracy on MNIST, CIFAR-10, and CIFAR-100 datasets.

Demonstrates mitigation of barren plateaus in 15-qubit, 6-layer circuits.

Provides a generalized training approach applicable to various QNN architectures.

Abstract

Training deep quantum neural networks (QNNs) for image classification is notoriously difficult due to vanishing gradients (barren plateaus) and limited nonlinearity in purely unitary circuits. We propose a novel gradient-free surrogate-driven framework combined with mid-circuit measurement and reset of ancillary qubits to induce effective nonunitarity. Our approach uses a classical neural surrogate to predict measurement outcomes from circuit parameters to avoid direct gradients. Theoretical results prove that bypassing quantum gradients mitigates plateau issues. Experiments on MNIST, CIFAR-10, and CIFAR-100 with 15-qubit, 6-layer circuits using four resettable ancillas demonstrate superior accuracy compared to direct-gradient QNNs and classical baselines. Our method also serves as a potential for a generalized training framework applicable to various QNN architectures beyond image classification.