DiffQ: Unified Parameter Initialization for Variational Quantum Algorithms via Diffusion Models

2025-09-21

AI Summary

Variational quantum algorithms (VQAs) suffer from optimization difficulties on NISQ devices due to poor parameter initialization. Existing machine learning–based initialization methods exhibit limited generalizability, relying heavily on single-task and small-sample settings. This work pioneers a generative learning perspective for VQA parameter initialization and proposes DiffQ—a unified initialization framework grounded in denoising diffusion probabilistic models (DDPMs). To overcome the constraints of small-scale, task-specific benchmarks, we construct the first large-scale, cross-domain benchmark dataset comprising 15,085 instances spanning three quantum domains and five distinct VQA tasks. Extensive experiments demonstrate that DiffQ substantially improves training efficiency: it reduces initial loss by up to 8.95× and decreases convergence steps by up to 23.4%, consistently outperforming all existing baselines across diverse VQA applications.

Technology Category

• Machine Learning: Quantum Machine Learning
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Variational Quantum Algorithms (VQAs) are widely used in the noisy intermediate-scale quantum (NISQ) era, but their trainability and performance depend critically on initialization parameters that shape the optimization landscape. Existing machine learning-based initializers achieve state-of-the-art results yet remain constrained to single-task domains and small datasets of only hundreds of samples. We address these limitations by reformulating VQA parameter initialization as a generative modeling problem and introducing DiffQ, a parameter initializer based on the Denoising Diffusion Probabilistic Model (DDPM). To support robust training and evaluation, we construct a dataset of 15,085 instances spanning three domains and five representative tasks. Experiments demonstrate that DiffQ surpasses baselines, reducing initial loss by up to 8.95 and convergence steps by up to 23.4%.

Problem

• Improving VQA trainability through diffusion-based parameter initialization
• Overcoming single-task limitations in quantum algorithm initialization
• Enhancing optimization landscape with generative modeling for VQAs

Innovation

• Uses diffusion models for parameter initialization
• Reformulates initialization as generative modeling problem
• Constructs multi-domain dataset for robust evaluation