Noise-injected analog Ising machines enable ultrafast statistical sampling and machine learning

TL;DR

This paper introduces a noise-injection method in analog Ising machines to enable ultrafast statistical sampling, demonstrating potential for efficient machine learning and neural network training beyond traditional optimization tasks.

Contribution

The paper presents a universal noise-injection approach in analog Ising machines, enabling fast statistical sampling and neural network training with accuracy comparable to software methods.

Findings

Ising machines can perform Boltzmann distribution sampling accurately.

Sampling speed is orders-of-magnitude faster than software methods.

The approach enables efficient machine learning applications.

Abstract

Ising machines are a promising non-von-Neumann computational concept for neural network training and combinatorial optimization. However, while various neural networks can be implemented with Ising machines, their inability to perform fast statistical sampling makes them inefficient for training neural networks compared to digital computers. Here, we introduce a universal concept to achieve ultrafast statistical sampling with analog Ising machines by injecting noise. With an opto-electronic Ising machine, we experimentally demonstrate that this can be used for accurate sampling of Boltzmann distributions and for unsupervised training of neural networks, with equal accuracy as software-based training. Through simulations, we find that Ising machines can perform statistical sampling orders-of-magnitudes faster than software-based methods. This enables the use of Ising machines beyond combinatorial optimization and makes them into efficient tools for machine learning and other applications.