서브메뉴
검색
Quantum Machine Learning in Noisy Intermediate-Scale Quantum Era
Quantum Machine Learning in Noisy Intermediate-Scale Quantum Era
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202104812
- ISBN
- 9798293894499
- DDC
- 530.1
- 저자명
- Hu, Fangjun.
- 서명/저자
- Quantum Machine Learning in Noisy Intermediate-Scale Quantum Era
- 발행사항
- [Sl] : Princeton University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 239 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Tureci, Hakan E.
- 학위논문주기
- Thesis (Ph.D.)--Princeton University, 2025.
- 초록/해제
- 요약With the rapid advancement of quantum technologies, the application of quantum computing to machine learning (ML) tasks has attracted growing interest, especially in the Noisy Intermediate-Scale Quantum (NISQ) era. The emerging field of Quantum Machine Learning (QML) explores how quantum resources can benefit ML, yet concrete advantages over classical methods on classical data remain elusive. In practice, QML faces several major obstacles: barren plateaus - a phenomenon where the loss function gradient vanishes - hinder effective training, while quantum shot noise - uncertainty from finite sampling - limits both training and generalization accuracy. Additionally, the limited coherence time of NISQ devices poses challenges for learning on temporal or streaming data. Among various QML approaches, Quantum Reservoir Computing (QRC) has emerged as a promising alternative. Inspired by classical recurrent neural networks and quantum kernel methods, QRC circumvents the need for extensive parameter training of deep quantum circuits, thereby avoiding barren plateaus. However, QRC remains affected by quantum noise and decoherence.This thesis introduces theoretical frameworks - Resolvable Expressive Capacity and Eigentask Analysis - which rigorously quantify the expressive power of QRC under shot noise. By identifying the noise-resilient features of the function space of QRC, this framework provides practical guidelines for enhancing generalization in noisy quantum systems. It also reveals deep connections between QML, quantum metrology, and quantum dynamics, opening new avenues for interdisciplinary research. Furthermore, I present a hardware-compatible implementation, NISQ Reservoir Computing (NISQRC), which uses partial measurements and deterministic resets to realize QRC with a finite temporal memory that persists indefinitely. Finally, I explore a novel QML optimization strategy - Reservoir Gradient Descent - that provides a surrogate loss landscape derived from the reservoir in the presence of shot noise, by fully utilizing the convexity in output layers of QML. Most of these theoretical investigations have been experimentally validated on state-of-the-art superconducting quantum platforms.These approaches enable efficient parameter training, stronger generalization ability, and longer memory persistence. They provide powerful tools to more robust and higher-performance QML algorithms suited for current and near-term quantum hardware.
- 일반주제명
- Quantum physics
- 일반주제명
- Computer science
- 일반주제명
- Computational physics
- 키워드
- Quantum sensing
- 기타저자
- Princeton University Electrical and Computer Engineering
- 기본자료저록
- Dissertations Abstracts International. 87-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017358942
■00520260202104812
■006m o d
■007cr#unu||||||||
■020 ▼a9798293894499
■035 ▼a(MiAaPQ)AAI32167568
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a530.1
■1001 ▼aHu, Fangjun.▼0(orcid)0000-0003-1955-3724
■24510▼aQuantum Machine Learning in Noisy Intermediate-Scale Quantum Era
■260 ▼a[Sl]▼bPrinceton University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a239 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Tureci, Hakan E.
■5021 ▼aThesis (Ph.D.)--Princeton University, 2025.
■520 ▼aWith the rapid advancement of quantum technologies, the application of quantum computing to machine learning (ML) tasks has attracted growing interest, especially in the Noisy Intermediate-Scale Quantum (NISQ) era. The emerging field of Quantum Machine Learning (QML) explores how quantum resources can benefit ML, yet concrete advantages over classical methods on classical data remain elusive. In practice, QML faces several major obstacles: barren plateaus - a phenomenon where the loss function gradient vanishes - hinder effective training, while quantum shot noise - uncertainty from finite sampling - limits both training and generalization accuracy. Additionally, the limited coherence time of NISQ devices poses challenges for learning on temporal or streaming data. Among various QML approaches, Quantum Reservoir Computing (QRC) has emerged as a promising alternative. Inspired by classical recurrent neural networks and quantum kernel methods, QRC circumvents the need for extensive parameter training of deep quantum circuits, thereby avoiding barren plateaus. However, QRC remains affected by quantum noise and decoherence.This thesis introduces theoretical frameworks - Resolvable Expressive Capacity and Eigentask Analysis - which rigorously quantify the expressive power of QRC under shot noise. By identifying the noise-resilient features of the function space of QRC, this framework provides practical guidelines for enhancing generalization in noisy quantum systems. It also reveals deep connections between QML, quantum metrology, and quantum dynamics, opening new avenues for interdisciplinary research. Furthermore, I present a hardware-compatible implementation, NISQ Reservoir Computing (NISQRC), which uses partial measurements and deterministic resets to realize QRC with a finite temporal memory that persists indefinitely. Finally, I explore a novel QML optimization strategy - Reservoir Gradient Descent - that provides a surrogate loss landscape derived from the reservoir in the presence of shot noise, by fully utilizing the convexity in output layers of QML. Most of these theoretical investigations have been experimentally validated on state-of-the-art superconducting quantum platforms.These approaches enable efficient parameter training, stronger generalization ability, and longer memory persistence. They provide powerful tools to more robust and higher-performance QML algorithms suited for current and near-term quantum hardware.
■590 ▼aSchool code: 0181.
■650 4▼aQuantum physics
■650 4▼aComputer science
■650 4▼aComputational physics
■653 ▼aPhysical learning
■653 ▼aQuantum computing
■653 ▼aQuantum Machine Learning
■653 ▼aQuantum sensing
■690 ▼a0599
■690 ▼a0984
■690 ▼a0800
■690 ▼a0216
■71020▼aPrinceton University▼bElectrical and Computer Engineering.
■7730 ▼tDissertations Abstracts International▼g87-04B.
■790 ▼a0181
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17358942▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
