서브메뉴
검색
Learning Transferable Representations Across Domains- [electronic resource]
![Learning Transferable Representations Across Domains - [electronic resource]](/Users/Baul/Images/book.png)
Learning Transferable Representations Across Domains- [electronic resource]
상세정보
- 자료유형
- 학위논문파일 국외
- 최종처리일시
- 20240214095926
- ISBN
- 9798380367844
- DDC
- 621.3
- 저자명
- Yue, Xiangyu.
- 서명/저자
- Learning Transferable Representations Across Domains - [electronic resource]
- 발행사항
- [S.l.]: : University of California, Berkeley., 2022
- 발행사항
- Ann Arbor : : ProQuest Dissertations & Theses,, 2022
- 형태사항
- 1 online resource(149 p.)
- 주기사항
- Source: Dissertations Abstracts International, Volume: 85-03, Section: B.
- 주기사항
- Advisor: Sangiovanni-Vincentelli, Alberto.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2022.
- 사용제한주기
- This item must not be sold to any third party vendors.
- 초록/해제
- 요약Deep neural networks have achieved great success in learning representations on a given dataset. However, in many cases, the learned representations are dataset-dependent and cannot be transferred to datasets with different distributions, even for the same task. How to deal with domain shift is crucial to improve the generalization capability of models. Domain adaptation offers a potential solution, allowing us to transfer networks from a source domain with abundant labels onto target domains with only limited or no labels.In this dissertation, I will present the many ways that we can learn transferable representations under different scenarios, including 1) when the source domain has only limited labels, even only one label per class, 2) when there are multiple labeled source domains, 3) when there are multiple unseen unlabeled target domains. These approaches are general across different data modalities (e.g. vision and language) and can be easily combined to solve other similar domain transfer settings (e.g. adapting from multiple sources with limited labels), enabling models to generalize beyond the source domains. Many of the works transfer knowledge from simulation data to real-world data in order to alleviate the need for expensive manual annotations. Finally, I present our pioneering work on building a LiDAR point cloud simulator, which has further enabled a large amount of domain adaptation work on LiDAR point cloud segmentation adaptation.
- 일반주제명
- Electrical engineering.
- 일반주제명
- Computer science.
- 키워드
- Datasets
- 기타저자
- University of California, Berkeley Electrical Engineering & Computer Sciences
- 기본자료저록
- Dissertations Abstracts International. 85-03B.
- 기본자료저록
- Dissertation Abstract International
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008240612s2022 us |||||||||||||||c||eng d■001000016931146
■00520240214095926
■006m o d
■007cr#unu||||||||
■020 ▼a9798380367844
■035 ▼a(MiAaPQ)AAI29327524
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a621.3
■1001 ▼aYue, Xiangyu.
■24510▼aLearning Transferable Representations Across Domains▼h[electronic resource]
■260 ▼a[S.l.]:▼bUniversity of California, Berkeley. ▼c2022
■260 1▼aAnn Arbor :▼bProQuest Dissertations & Theses, ▼c2022
■300 ▼a1 online resource(149 p.)
■500 ▼aSource: Dissertations Abstracts International, Volume: 85-03, Section: B.
■500 ▼aAdvisor: Sangiovanni-Vincentelli, Alberto.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2022.
■506 ▼aThis item must not be sold to any third party vendors.
■520 ▼aDeep neural networks have achieved great success in learning representations on a given dataset. However, in many cases, the learned representations are dataset-dependent and cannot be transferred to datasets with different distributions, even for the same task. How to deal with domain shift is crucial to improve the generalization capability of models. Domain adaptation offers a potential solution, allowing us to transfer networks from a source domain with abundant labels onto target domains with only limited or no labels.In this dissertation, I will present the many ways that we can learn transferable representations under different scenarios, including 1) when the source domain has only limited labels, even only one label per class, 2) when there are multiple labeled source domains, 3) when there are multiple unseen unlabeled target domains. These approaches are general across different data modalities (e.g. vision and language) and can be easily combined to solve other similar domain transfer settings (e.g. adapting from multiple sources with limited labels), enabling models to generalize beyond the source domains. Many of the works transfer knowledge from simulation data to real-world data in order to alleviate the need for expensive manual annotations. Finally, I present our pioneering work on building a LiDAR point cloud simulator, which has further enabled a large amount of domain adaptation work on LiDAR point cloud segmentation adaptation.
■590 ▼aSchool code: 0028.
■650 4▼aElectrical engineering.
■650 4▼aComputer science.
■653 ▼aDomain adaptation
■653 ▼aDeep neural networks
■653 ▼aDatasets
■653 ▼aLiDAR point cloud
■690 ▼a0544
■690 ▼a0984
■71020▼aUniversity of California, Berkeley▼bElectrical Engineering & Computer Sciences.
■7730 ▼tDissertations Abstracts International▼g85-03B.
■773 ▼tDissertation Abstract International
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2022
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T16931146▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
■980 ▼a202402▼f2024
