Communication-efficient Quantum Algorithm for Distributed Machine Learning

TL;DR

This paper introduces a quantum algorithm that significantly reduces communication costs for distributed machine learning tasks like least-square fitting and softmax regression, especially with large datasets.

Contribution

It presents a novel quantum algorithm that achieves lower communication complexity than classical and existing quantum methods for distributed ML problems.

Findings

Achieves $O(rac{ ext{log}_2(N)}{ ext{epsilon}})$ communication complexity

Provides quantum-accelerated estimation of inner products and Hamming distances

Offers communication advantages over classical algorithms for large datasets

Abstract

The growing demands of remote detection and increasing amount of training data make distributed machine learning under communication constraints a critical issue. This work provides a communication-efficient quantum algorithm that tackles two traditional machine learning problems, the least-square fitting and softmax regression problem, in the scenario where the data set is distributed across two parties. Our quantum algorithm finds the model parameters with a communication complexity of $O(\frac{\log_2(N)}{\epsilon})$, where $N$ is the number of data points and $\epsilon$ is the bound on parameter errors. Compared to classical algorithms and other quantum algorithms that achieve the same output task, our algorithm provides a communication advantage in the scaling with the data volume. The building block of our algorithm, the quantum-accelerated estimation of distributed inner product and Hamming distance, could be further applied to various tasks in distributed machine learning to accelerate communication.