서브메뉴
검색
Disentangled Representation Learning with Quantized Latents
Disentangled Representation Learning with Quantized Latents
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105610
- ISBN
- 9798265427823
- DDC
- 330
- 저자명
- Hsu, Kyle.
- 서명/저자
- Disentangled Representation Learning with Quantized Latents
- 발행사항
- [Sl] : Stanford University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 100 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Finn, Chelsea;Wu, Jiajun.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2025.
- 초록/해제
- 요약Disentangled representation learning seeks to identify the generative source variables underlying high-dimensional data. Despite its conceptual appeal, progress in this area has been limited by the lack of principled evaluation metrics, weak inductive biases, and the persistence of overly restrictive assumptions such as source independence. This thesis addresses these challenges through four key contributions. First, it introduces InfoMEC, a set of information-theoretic metrics for quantifying modularity, explicitness, and compactness in learned representations. Second, it proposes latent quantization as an architectural inductive bias for state-of-the-art disentanglement. Third, it presents Tripod, a model that synergistically integrates latent quantization with latent multiinformation regularization and latent functional interdependence regularization to further push the state-of-the-art. Finally, it broadens the theoretical foundations of disentanglement by replacing the assumption of source statistical independence with source range independence, a weaker condition under which modularity still emerges in the models studied in this thesis. Together, these contributions advance both the theory and practice of disentangled representation learning.
- 일반주제명
- Sparsity
- 일반주제명
- Decomposition
- 일반주제명
- Lattice theory
- 일반주제명
- Entropy
- 일반주제명
- Natural language
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360718
■00520260202105610
■006m o d
■007cr#unu||||||||
■020 ▼a9798265427823
■035 ▼a(MiAaPQ)AAI32316384
■035 ▼a(MiAaPQ)Stanforddp898pk5594
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a330
■1001 ▼aHsu, Kyle.
■24510▼aDisentangled Representation Learning with Quantized Latents
■260 ▼a[Sl]▼bStanford University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a100 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Finn, Chelsea;Wu, Jiajun.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2025.
■520 ▼aDisentangled representation learning seeks to identify the generative source variables underlying high-dimensional data. Despite its conceptual appeal, progress in this area has been limited by the lack of principled evaluation metrics, weak inductive biases, and the persistence of overly restrictive assumptions such as source independence. This thesis addresses these challenges through four key contributions. First, it introduces InfoMEC, a set of information-theoretic metrics for quantifying modularity, explicitness, and compactness in learned representations. Second, it proposes latent quantization as an architectural inductive bias for state-of-the-art disentanglement. Third, it presents Tripod, a model that synergistically integrates latent quantization with latent multiinformation regularization and latent functional interdependence regularization to further push the state-of-the-art. Finally, it broadens the theoretical foundations of disentanglement by replacing the assumption of source statistical independence with source range independence, a weaker condition under which modularity still emerges in the models studied in this thesis. Together, these contributions advance both the theory and practice of disentangled representation learning.
■590 ▼aSchool code: 0212.
■650 4▼aSparsity
■650 4▼aDecomposition
■650 4▼aLattice theory
■650 4▼aEntropy
■650 4▼aNatural language
■690 ▼a0800
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360718▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
