서브메뉴
검색
Efficient Machine Learning for Intelligent Machines
Efficient Machine Learning for Intelligent Machines
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105416
- ISBN
- 9798293892679
- DDC
- 620
- 저자명
- Xu, Chenfeng.
- 서명/저자
- Efficient Machine Learning for Intelligent Machines
- 발행사항
- [Sl] : University of California, Berkeley, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 272 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Tomizuka, Masayoshi;Keutzer, Kurt.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2025.
- 초록/해제
- 요약The rapid advancement of machine learning (ML) has been driven by the exponential scaling of computational resources, model complexity, and data availability. Modern ML systems-ranging from large language models (LLMs) to vision and robotics systems-now exhibit remarkable capabilities such as in-context learning, multi-modal generation, robust robotic perception and planning. However, this progress has come at the cost of increasing computational and data demands that far exceed the practical constraints of real-world intelligent machines. These constraints are particularly acute in settings such as autonomous driving, robotics, and edge computing.This dissertation tackles the core challenges of computational and data efficiency in intelligent machines through a unified framework grounded in the structural properties of hardware, models, and data. The contributions are organized into three main parts. The first part enhances computational efficiency by introducing compact and structured data representations, including efficient tokenization schemes and geometric abstractions for point cloud and multi-view video processing. The second part proposes new methods for reorganizing model and system-level computation-such as spatially-aware convolutions, low-rank decomposition for large language models, and throughput-optimized diffusion pipelines. The third part addresses data efficiency by leveraging structured priors and generative data engines, enabling few-shot learning via 2D-to-3D transfer, disentangled generative training, and viewpoint-conditioned augmentation for robotic manipulation. Collectively, these contributions aim to advance the frontiers of real-world machine learning by developing systems that are not only more capable but also significantly more efficient and deployable across a wide range of intelligent machine applications.
- 일반주제명
- Engineering
- 일반주제명
- Robotics
- 일반주제명
- Mechanical engineering
- 기타저자
- University of California, Berkeley Mechanical Engineering
- 기본자료저록
- Dissertations Abstracts International. 87-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360271
■00520260202105416
■006m o d
■007cr#unu||||||||
■020 ▼a9798293892679
■035 ▼a(MiAaPQ)AAI32236321
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a620
■1001 ▼aXu, Chenfeng.
■24510▼aEfficient Machine Learning for Intelligent Machines
■260 ▼a[Sl]▼bUniversity of California, Berkeley▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a272 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Tomizuka, Masayoshi;Keutzer, Kurt.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2025.
■520 ▼aThe rapid advancement of machine learning (ML) has been driven by the exponential scaling of computational resources, model complexity, and data availability. Modern ML systems-ranging from large language models (LLMs) to vision and robotics systems-now exhibit remarkable capabilities such as in-context learning, multi-modal generation, robust robotic perception and planning. However, this progress has come at the cost of increasing computational and data demands that far exceed the practical constraints of real-world intelligent machines. These constraints are particularly acute in settings such as autonomous driving, robotics, and edge computing.This dissertation tackles the core challenges of computational and data efficiency in intelligent machines through a unified framework grounded in the structural properties of hardware, models, and data. The contributions are organized into three main parts. The first part enhances computational efficiency by introducing compact and structured data representations, including efficient tokenization schemes and geometric abstractions for point cloud and multi-view video processing. The second part proposes new methods for reorganizing model and system-level computation-such as spatially-aware convolutions, low-rank decomposition for large language models, and throughput-optimized diffusion pipelines. The third part addresses data efficiency by leveraging structured priors and generative data engines, enabling few-shot learning via 2D-to-3D transfer, disentangled generative training, and viewpoint-conditioned augmentation for robotic manipulation. Collectively, these contributions aim to advance the frontiers of real-world machine learning by developing systems that are not only more capable but also significantly more efficient and deployable across a wide range of intelligent machine applications.
■590 ▼aSchool code: 0028.
■650 4▼aEngineering
■650 4▼aRobotics
■650 4▼aMechanical engineering
■653 ▼aEfficient machine learning
■653 ▼aIntelligent machines
■653 ▼aRobotic perception
■690 ▼a0537
■690 ▼a0771
■690 ▼a0800
■690 ▼a0548
■71020▼aUniversity of California, Berkeley▼bMechanical Engineering.
■7730 ▼tDissertations Abstracts International▼g87-04B.
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360271▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
