본문

서브메뉴

Advancing Efficiency and Trustworthiness: From Computer Vision to Multimodal Large Language Models
Advancing Efficiency and Trustworthiness: From Computer Vision to Multimodal Large Languag...
Advancing Efficiency and Trustworthiness: From Computer Vision to Multimodal Large Language Models

Detailed Information

자료유형  
 학위논문 서양
최종처리일시  
20260311091529.5
ISBN  
9798270231293
DDC  
000
저자명  
Li, Zhangheng
서명/저자  
Advancing Efficiency and Trustworthiness: From Computer Vision to Multimodal Large Language Models / Zhangheng Li
발행사항  
[Sl] : The University of Texas at Austin, 2025
형태사항  
1 electronic resource (191 pages)
주기사항  
Source: Dissertations Abstracts International, Volume: 87-06, Section: B.
주기사항  
Advisors: Wang, Zhangyang Committee members: Chen, Tianlong; Ding, Ying; de Veciana, Gustavo; Ghosh, Joydeep.
학위논문주기  
- Ph.D. : The University of Texas at Austin, 2025.
초록/해제  
요약Deep learning has revolutionized artificial intelligence, but its growing computational demands and trustworthiness concerns present significant challenges. This thesis investigates the intersection of efficiency and trustworthiness in deep neural networks, proposing novel approaches that advance both dimensions simultaneously. First, we explore techniques for neural network efficiency through pruning and quantization. We introduce mixed-precision quantization tickets that unify pruning and quantization paradigms, and develop sparse cocktail training to produce multiple sparse subnetworks with diverse sparsity characteristics from a single training process. These methods achieve substantial computational savings while maintaining competitive performance. Second, we examine trustworthiness across multiple dimensions including robustness, privacy, fairness, and ethics. We present a comprehensive framework for evaluating compressed models' trustworthiness, revealing that quantization generally preserves trustworthiness better than pruning at similar compression rates. Additionally, we demonstrate how pruning can enhance certified robustness verification through stability-based pruning techniques. Third, we investigate the synergies between efficiency and trustworthiness, showing how these properties can mutually reinforce each other when properly designed. We leverage sparse transfer learning to accelerate certified robustness verification while maintaining robustness guarantees. Finally, we extend our research to multimodal learning, developing efficient and trustworthy multimodal agents. We introduce a universal UI understanding model supporting multiple platforms with adaptive high-resolution processing, and address privacy risks in diffusion models and multimodal large language models through novel unlearning methods and vulnerability assessments. Collectively, this research advances the state-of-the-art in both efficient and trustworthy deep learning, providing practical approaches for developing models that are simultaneously resource-efficient and reliable across multiple dimensions of trustworthiness.
언어주기  
English
일반주제명  
Computer engineering
일반주제명  
Computer science
일반주제명  
Information technology
키워드  
Trustworthiness
키워드  
Computer vision
키워드  
Large language models
키워드  
Deep learning
키워드  
Robustness
기타저자  
The University of Texas at Austin Electrical and Computer Engineering
기본자료저록  
Dissertations Abstracts International. 87-06B.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260311s2025        us                                    eng  d
■001000017361182
■00520260311091529.5
■006m          o    d                
■007cr|nu||||||||
■020    ▼a9798270231293
■040    ▼aMiAaPQD▼beng▼cMiAaPQD▼erda
■082    ▼a000
■1001  ▼aLi,  Zhangheng▼eauthor.
■24510▼aAdvancing  Efficiency  and  Trustworthiness:  From  Computer  Vision  to  Multimodal  Large  Language  Models  ▼cZhangheng  Li
■260    ▼a[Sl]▼bThe  University  of  Texas  at  Austin▼c2025
■264  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2025
■300    ▼a1  electronic  resource  (191  pages)
■336    ▼atext▼btxt▼2rdacontent
■337    ▼acomputer▼bc▼2rdamedia
■338    ▼aonline  resource▼bcr▼2rdacarrier
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  87-06,  Section:  B.
■500    ▼aAdvisors:  Wang,  Zhangyang    Committee  members:  Chen,  Tianlong;  Ding,  Ying;  de  Veciana,  Gustavo;  Ghosh,  Joydeep.
■5021  ▼bPh.D.▼cThe  University  of  Texas  at  Austin▼d2025.
■520    ▼aDeep  learning  has  revolutionized  artificial  intelligence,  but  its  growing  computational  demands  and  trustworthiness  concerns  present  significant  challenges.  This  thesis  investigates  the  intersection  of  efficiency  and  trustworthiness  in  deep  neural  networks,  proposing  novel  approaches  that  advance  both  dimensions  simultaneously.                                                First,  we  explore  techniques  for  neural  network  efficiency  through  pruning  and  quantization.  We  introduce  mixed-precision  quantization  tickets  that  unify  pruning  and  quantization  paradigms,  and  develop  sparse  cocktail  training  to  produce  multiple  sparse  subnetworks  with  diverse  sparsity  characteristics  from  a  single  training  process.  These  methods  achieve  substantial  computational  savings  while  maintaining  competitive  performance.                                                Second,  we  examine  trustworthiness  across  multiple  dimensions  including  robustness,  privacy,  fairness,  and  ethics.  We  present  a  comprehensive  framework  for  evaluating  compressed  models'  trustworthiness,  revealing  that  quantization  generally  preserves  trustworthiness  better  than  pruning  at  similar  compression  rates.  Additionally,  we  demonstrate  how  pruning  can  enhance  certified  robustness  verification  through  stability-based  pruning  techniques.                                                Third,  we  investigate  the  synergies  between  efficiency  and  trustworthiness,  showing  how  these  properties  can  mutually  reinforce  each  other  when  properly  designed.  We  leverage  sparse  transfer  learning  to  accelerate  certified  robustness  verification  while  maintaining  robustness  guarantees.                                                Finally,  we  extend  our  research  to  multimodal  learning,  developing  efficient  and  trustworthy  multimodal  agents.  We  introduce  a  universal  UI  understanding  model  supporting  multiple  platforms  with  adaptive  high-resolution  processing,  and  address  privacy  risks  in  diffusion  models  and  multimodal  large  language  models  through  novel  unlearning  methods  and  vulnerability  assessments.                                                Collectively,  this  research  advances  the  state-of-the-art  in  both  efficient  and  trustworthy  deep  learning,  providing  practical  approaches  for  developing  models  that  are  simultaneously  resource-efficient  and  reliable  across  multiple  dimensions  of  trustworthiness.
■546    ▼aEnglish
■590    ▼aSchool  code:  0227
■650  4▼aComputer  engineering
■650  4▼aComputer  science
■650  4▼aInformation  technology
■653    ▼aTrustworthiness
■653    ▼aComputer  vision
■653    ▼aLarge  language  models
■653    ▼aDeep  learning
■653    ▼aRobustness
■7102  ▼aThe  University  of  Texas  at  Austin▼bElectrical  and  Computer  Engineering.▼edegree  granting  institution.
■7201  ▼aWang,  Zhangyang▼edegree  supervisor.
■7730  ▼tDissertations  Abstracts  International▼g87-06B.
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17361182▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

Preview

Export

ChatGPT Discussion

AI Recommended Related Books


    New Books MORE
    Statistics for the past 3 years. Go to brief

    Подробнее информация.

    • Бронирование
    • не существует
    • моя папка
    • Первый запрос зрения
    • Non-Book Loan Application
    • Nighttime Book Loan Application
    материал
    Reg No. Количество платежных Местоположение статус Ленд информации
    TF17808 전자도서 대출가능 My Folder 부재도서신고 비도서대출신청 야간 도서대출신청

    * Бронирование доступны в заимствований книги. Чтобы сделать предварительный заказ, пожалуйста, нажмите кнопку бронирование

    Books borrowed together with this book

    Related Popular Books

    Available after logging in.