MLQM: Machine Learning Approach for Accelerating Optimal Qubit Mapping

TL;DR

MLQM introduces a machine learning-based method to accelerate optimal qubit mapping in quantum circuits, significantly improving speed and space efficiency over existing techniques.

Contribution

The paper presents a novel ML-driven approach with global and local search space pruning, plus data augmentation, to enhance quantum circuit mapping efficiency.

Findings

Achieves an average 1.79x speed-up in solving time.

Demonstrates a 22% reduction in space complexity.

Outperforms state-of-the-art methods in efficiency.

Abstract

Quantum circuit mapping is a critical process in quantum computing that involves adapting logical quantum circuits to adhere to hardware constraints, thereby generating physically executable quantum circuits. Current quantum circuit mapping techniques, such as solver-based methods, often encounter challenges related to slow solving speeds due to factors like redundant search iterations. Regarding this issue, we propose a machine learning approach for accelerating optimal qubit mapping (MLQM). First, the method proposes a global search space pruning scheme based on prior knowledge and machine learning, which in turn improves the solution efficiency. Second, to address the limited availability of effective samples in the learning task, MLQM introduces a novel data augmentation and refinement scheme, this scheme enhances the size and diversity of the quantum circuit dataset by exploiting gate allocation and qubit rearrangement. Finally, MLQM also further improves the solution efficiency by pruning the local search space, which is achieved through an adaptive dynamic adjustment mechanism of the solver variables. Compared to state-of-the-art qubit mapping approaches, MLQM achieves optimal qubit mapping with an average solving speed-up ratio of 1.79 and demonstrates an average advantage of 22% in terms of space complexity.