Approaches for utilizing quantum computing with limited resources (本文)
Year: 2024
Type: dissertation
Abstract: Quantum computers are expected to solve certain problems more efficiently than conventional computers by leveraging the laws of quantum mechanics.With rapid advancements in quantum computer development, these devices have become widely accessible through cloud platforms.However, current quantum computers are constrained by limited resources.They only have a few hundred noisy quantum bits (qubits) available, which restricts the range of algorithms that can be run on these devices.Those working with current quantum computers are faced with the problem of utilizing a restricted number of noisy qubits with limited access to hardware parameters.This thesis aims to solve the following problems: the shortage of a simple method for benchmarking qubits of quantum computing devices, and the limited understanding of the noise robustness of optimization algorithms using Quantum Random Access Code (QRAC).Solving these problems provides tools and insight to facilitate the utilization of devices with limited resources.The first challenge is the need to select stable qubits from quantum computers.Especially with cloudbased quantum computers that can only be accessed remotely, the only information that is guaranteed access is the input and output.To address this challenge, we propose a method for evaluating the stability of qubits of a quantum computing device based only on the input and output.We focus on the task of quantum random number generation, an inherently quantum mechanical task for quantum computers that leverages the Born rule.We find that existing statistical tests for random number generators (RNGs) cannot be directly applied to assess the stability of the qubits of current quantum computing devices because their qubits do not satisfy the unbiased assumption.To address this issue, the proposed method uses a polyline approximation algorithm on the cumulative distribution of the output, allowing for a pure evaluation of stability.We demonstrate through experiments on a real device that the method can detect variations in stability of qubits over time.The second challenge is the need for noise robust algorithms.Real devices are affected by noise, whose intensity varies over time.To consistently obtain useful results, it is vital to choose algorithms that are robust to noise.In this study, we focus on the noise robustness of a particular combinatorial optimization algorithm referred to as Quantum Random Access Optimization (QRAO).Combinatorial optimization is a task with a wide range of applications, and it is one of the promising applications for quantum computers.QRAO is expected to efficiently solve Quantum Unconstrained Binary Optimization (QUBO) problems with fewer qubits using Quantum Random Access Code (QRAC), but its noise robustness was not sufficiently understood.In particular, the measurement process called quantum state rounding, which extracts classical solutions from quantum states, had unclear effects on the quality of the solutions under noise.In this study, we use the (3,1)-QRAC Hamiltonian, which encodes up to three classical bits per qubit, and numerically demonstrate its higher noise robustness compared to the Ising Hamiltonian, with the MaxCut problem as an example.We also analyze the mechanisms behind the noise robustness and derive the number of measurements required to accurately decode classical bits with a given precision.The findings of this research provide a method for easily evaluating the stability of noisy qubits and offer insight into the noise-robustness of combinatorial optimization using QRAC.These contributions are expected to help address key challenges for utilizing quantum computers with limited resources.A4 The relationship between the normal distribution 𝑁 (100 × 0.5, √ 100 × 0.5) and its cumulative distribution function Φ(𝑥).The area shaded in blue corresponds to the value of the cumulative distribution function Φ(𝑥) for 𝑥 = 60. . . . . . . . . . . . . . . . . . . .96 A5 The cumulative distribution function Φ(𝑥) corresponding to 𝑁 (100 × 0.5, √ 100 × 0.5). .96 A6 The visual representation of the two-sided P-value of 𝑥 ∼ 𝑁 (0, 1) for 𝑥 = 2. . . . . . . .97 A7 The 𝜒 2 distribution of 𝑘 = 1, 2, 3, 4, and 5 degrees of freedom. . . . . . . . . . . . . . .97 A8 The meaning of the P-value of the 𝜒 2 test, where the number of classes is 10 and the observed 𝜒 2 test statistic value is 7.60. . . . . . . . . . . . . . . . . . . . . . . . . . . .98This thesis is the culmination of the support and encouragement of many individuals and institutions.
Authors: タムラ, ケンタロウ, Tamura, Kentaro, 田村, 賢太郎
Institution: Keio University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Keio University
OpenAlex ID: W7054849630
