WSBD: Freezing-Based Optimizer for Quantum Neural Networks

2026-02-11

AI Summary

Training quantum neural networks is often hindered by high computational costs and the barren plateau problem. To address these challenges, this work proposes the Weighted Stochastic Block Descent (WSBD) optimizer, which innovatively integrates gradient-derived importance scoring, a parameter-level dynamic freezing mechanism, and stochastic block-wise optimization. This approach significantly reduces the number of forward passes while preserving model expressivity. Theoretical analysis provides formal convergence guarantees, and empirical results demonstrate that WSBD achieves an average 63.9% faster convergence than Adam on ground-state energy estimation tasks, with performance gains amplifying as network size increases.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Optimization
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Abstract

The training of Quantum Neural Networks (QNNs) is hindered by the high computational cost of gradient estimation and the barren plateau problem, where optimization landscapes become intractably flat. To address these challenges, we introduce Weighted Stochastic Block Descent (WSBD), a novel optimizer with a dynamic, parameter-wise freezing strategy. WSBD intelligently focuses computational resources by identifying and temporarily freezing less influential parameters based on a gradient-derived importance score. This approach significantly reduces the number of forward passes required per training step and helps navigate the optimization landscape more effectively. Unlike pruning or layer-wise freezing, WSBD maintains full expressive capacity while adapting throughout training. Our extensive evaluation shows that WSBD converges on average 63.9% faster than Adam for the popular ground-state-energy problem, an advantage that grows with QNN size. We provide a formal convergence proof for WSBD and show that parameter-wise freezing outperforms traditional layer-wise approaches in QNNs. Project page: https://github.com/Damrl-lab/WSBD-Stochastic-Freezing-Optimizer.

Problem

• Quantum Neural Networks
• barren plateau
• gradient estimation
• optimization landscape
• computational cost

Innovation

• Quantum Neural Networks
• Parameter-wise Freezing
• Barren Plateau
• Gradient Estimation
• Stochastic Optimization