Assessment of image generation by quantum annealer

Area of Science:

• Quantum Computing
• Machine Learning
• Statistical Physics

Background:

• Quantum annealing, initially for optimization, is explored for sampling and Boltzmann machine learning.
• D-Wave Systems offers quantum annealing hardware, but noise and environmental factors affect performance.
• Previous work compared quantum annealer sampling to classical methods using standard distance metrics.

Purpose of the Study:

• To evaluate quantum annealer performance as a generative model from a novel perspective.
• To assess its effectiveness in Boltzmann machine learning beyond standard distribution comparisons.

Main Methods:

• Utilized a quantum annealer for generative model training in Boltzmann machine learning.
• Employed a neural network discriminator, trained on an a priori dataset, for performance evaluation.
• Compared quantum annealer performance against classical approaches and ideal quantum annealing.

Main Results:

• Quantum annealer demonstrated superior performance in training generative models for Boltzmann machine learning compared to classical methods.
• Remanent quantum fluctuations in the hardware negatively impacted the quality of generated data.
• Generated image quality from the quantum annealer was inferior to ideal quantum annealing and classical Monte Carlo sampling.

Conclusions:

• Quantum annealers show potential as fast samplers and effective tools for Boltzmann machine learning.
• Hardware limitations, specifically remanent quantum fluctuations, currently hinder optimal generative model performance.
• Further research is needed to mitigate noise and improve data generation quality in quantum annealers.