Eur. Phys. J. Plus (2022) 137: 491
https://doi.org/10.1140/epjp/s13360-022-02714-7

Regular Article

Search space pruning for quantum architecture search

¹ School of Electronic and Information Engineering, Foshan University, 528000, Foshan, China
² School of Mechatronic Engineering and Automation, Foshan University, 528000, Foshan, China
³ Guangxi Key Laboratory of Cryptography and Information Security, Guilin University of Electronic Technology, 541004, Guilin, China
⁴ College of Mathematics and Informatics, South China Agricultural University, 510642, Guangzhou, China

^a zhmihe@gmail.com

Received: 16 December 2021
Accepted: 11 April 2022
Published online: 21 April 2022

Abstract

Variational quantum algorithms (VQAs) have been proposed for most of the applications that researchers have envisioned for quantum computers and appear to be the most promising strategy for quantum advantage on NISQ devices. The performance of VQAs largely depends on the structure of the quantum circuit. Manual design of a quantum circuit is time-consuming and requires human expertise. Quantum architecture search (QAS) algorithms aim to automate the design of quantum circuits in VQAs using classical optimization algorithms. However, the search space of QAS algorithms increases exponentially with the number of quantum gates. It is difficult and requires a lot of computing resources to find out an optimal circuit in such a large space by classical optimization algorithms. In this paper, we propose a space pruning algorithm, which can largely reduce the search space by progressively removing unpromising candidate gates. We use two indicators to evaluate the candidate gates (i.e., the cost-based and rotation-based indicators), and simulation result shows that the rotation-based indicator achieves a better performance than the cost-based one when they are estimated after a small number of updates on the gate parameters. The proposed method is a preprocessing method and can be used to improve the efficiency of arbitrary QAS algorithms. Simulation results show that a state-of-the-art QAS algorithm converges faster and achieves a lower cost in the pruned space than the original one.