서브메뉴
검색
A Perturbation Approach to Differential Privacy for Deep Learning Based Speech Processing
A Perturbation Approach to Differential Privacy for Deep Learning Based Speech Processing
Detailed Information
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260209102908
- ISBN
- 9798263393823
- DDC
- 004
- 저자명
- Yang, Chao-Han.
- 서명/저자
- A Perturbation Approach to Differential Privacy for Deep Learning Based Speech Processing
- 발행사항
- [Sl] : Georgia Institute of Technology, 2023
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2023
- 형태사항
- 135 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Lee, Chin-Hui.
- 학위논문주기
- Thesis (Ph.D.)--Georgia Institute of Technology, 2023.
- 초록/해제
- 요약Speech signals contain rich information that encompasses the characteristics of speakers and speaking environments. To deploy high-performance speech applications, deep neural network (DNN) based models are widely utilized and trained with speech data. However, the training set often contains personal information that needs to be protected. New data regulations (e.g., GDPR) require service providers to ensure necessary privacy protections through privacy measurements to avoid security threats from query-based malicious attacks. Differential privacy (DP) is a technique that provides identity protection with a mathematical definition for measuring privacy losses under query-based attacks. DP is built upon randomization processes for identity anonymization by setting up a privacy budget (溝). However, ensuring DP in DNN models often leads to a severe performance loss due to the data perturbation process for achieving a strong privacy budget.This dissertation aims to establish a perturbation-based learning framework for training with DNNs to maintain the high performance of speech models while satisfying DP requirements. We develop specific mechanisms for signal distortion by adding bounded noise to the speech training data. These perturbations can be analyzed through statistical measurements and estimations of the deployed models. Our theoretical studies have used both Laplace and Gaussian noise to guarantee the 溝 privacy budget under a bounded maxdivergence measurement with ensemble learning. We further leverage upon the properties of Lipchitz continuity of DNNs to provide model robustness estimation under different 溝 budgets. Speech recognition in both isolated speech commands and large vocabulary continuous speech settings has been utilized to corroborate our theorems.We have advanced the proposed framework into popular speech and acoustic modeling tasks and conducted preliminary investigations to explore the trade-off between ensemble model performance and DP budgets. Our findings show that even with state-of-the-art speaker anonymization techniques, speaker identity information can still be leaked without DP-based data extraction or training methods. Finally, we have incorporated decentralized and federated training pipelines of deep neural networks (DNNs) into the proposed privacypreserving speech processing mechanism to investigate potential solutions for on-device or cloud-based speech applications for end-users.
- 일반주제명
- Data integrity
- 일반주제명
- Privacy
- 일반주제명
- Neural networks
- 일반주제명
- Computer science
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260203s2023 us c eng d■001000017365983
■00520260209102908
■006m o d
■007cr#unu||||||||
■020 ▼a9798263393823
■035 ▼a(MiAaPQ)AAI32315763
■035 ▼a(MiAaPQ)GeorgiaTech75094
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aYang, Chao-Han.
■24512▼aA Perturbation Approach to Differential Privacy for Deep Learning Based Speech Processing
■260 ▼a[Sl]▼bGeorgia Institute of Technology▼c2023
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2023
■300 ▼a135 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Lee, Chin-Hui.
■5021 ▼aThesis (Ph.D.)--Georgia Institute of Technology, 2023.
■520 ▼aSpeech signals contain rich information that encompasses the characteristics of speakers and speaking environments. To deploy high-performance speech applications, deep neural network (DNN) based models are widely utilized and trained with speech data. However, the training set often contains personal information that needs to be protected. New data regulations (e.g., GDPR) require service providers to ensure necessary privacy protections through privacy measurements to avoid security threats from query-based malicious attacks. Differential privacy (DP) is a technique that provides identity protection with a mathematical definition for measuring privacy losses under query-based attacks. DP is built upon randomization processes for identity anonymization by setting up a privacy budget (溝). However, ensuring DP in DNN models often leads to a severe performance loss due to the data perturbation process for achieving a strong privacy budget.This dissertation aims to establish a perturbation-based learning framework for training with DNNs to maintain the high performance of speech models while satisfying DP requirements. We develop specific mechanisms for signal distortion by adding bounded noise to the speech training data. These perturbations can be analyzed through statistical measurements and estimations of the deployed models. Our theoretical studies have used both Laplace and Gaussian noise to guarantee the 溝 privacy budget under a bounded maxdivergence measurement with ensemble learning. We further leverage upon the properties of Lipchitz continuity of DNNs to provide model robustness estimation under different 溝 budgets. Speech recognition in both isolated speech commands and large vocabulary continuous speech settings has been utilized to corroborate our theorems.We have advanced the proposed framework into popular speech and acoustic modeling tasks and conducted preliminary investigations to explore the trade-off between ensemble model performance and DP budgets. Our findings show that even with state-of-the-art speaker anonymization techniques, speaker identity information can still be leaked without DP-based data extraction or training methods. Finally, we have incorporated decentralized and federated training pipelines of deep neural networks (DNNs) into the proposed privacypreserving speech processing mechanism to investigate potential solutions for on-device or cloud-based speech applications for end-users.
■590 ▼aSchool code: 0078.
■650 4▼aData integrity
■650 4▼aPrivacy
■650 4▼aNeural networks
■650 4▼aComputer science
■690 ▼a0800
■690 ▼a0984
■71020▼aGeorgia Institute of Technology.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0078
■791 ▼aPh.D.
■792 ▼a2023
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17365983▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
Preview
Export
ChatGPT Discussion
AI Recommended Related Books
Подробнее информация.
- Бронирование
- не существует
- моя папка
- Первый запрос зрения
- Non-Book Loan Application
- Nighttime Book Loan Application
Available after logging in.


