자료유형  

 학위논문

Control Number  

0017163727

International Standard Book Number  

9798342107174

Dewey Decimal Classification Number  

616.07

Main Entry-Personal Name  

Sagawa, Shiori.

Publication, Distribution, etc. (Imprint  

[S.l.] : Stanford University., 2024

Publication, Distribution, etc. (Imprint  

Ann Arbor : ProQuest Dissertations & Theses, 2024

Physical Description  

212 p.

General Note  

Source: Dissertations Abstracts International, Volume: 86-04, Section: B.

General Note  

Advisor: Liang, Percy.

Dissertation Note  

Thesis (Ph.D.)--Stanford University, 2024.

Summary, Etc.  

요약Machine learning models can unexpectedly fail in the wild due to distribution shifts:mismatches in data distributions between training and deployment. Distribution shifts are often unavoidable and pose significant reliability challenges in many real-world applications. For example, models can fail on certain subpopulations (e.g., language models can fail on non-English languages) and on new domains unseen during training (e.g., medical models can fail on new hospitals).In this thesis, we aim to build reliable machine learning models that are robust to distribution shifts in the wild. In the first part, we mitigate subpopulation shifts by developing methods that leverage distributionally robust optimization (DRO). These methods overcome the computational and statistical obstacles of applying DRO on modern neural networks and on real-world shifts. In the second part, to tackle domain shifts, we introduce WILDS, a benchmark of real-world shifts, and show that existing methods fail on WILDS even though they perform well on synthetic shifts from prior benchmarks. We then develop a method that successfully mitigates real-world domain shifts. We propose an alternative to domain invariance---a key principle behind the prior methods---to reflect the structure of real-world shifts, and our method achieves state-of-the-art performance on multiple WILDS datasets.Altogether, the algorithms developed in this thesis mitigate real-world distribution shifts by addressing key statistical and computational challenges of training robust models, while anchoring to and leveraging the structure of real-world distribution shifts. These algorithms successfully improve robustness to a wide range of distribution shifts in the wild, from subpopulation shifts in language modeling to domain shiftsin wildlife monitoring and histopathology to spurious correlations.

Subject Added Entry-Topical Term  

Histopathology.

Subject Added Entry-Topical Term  

Deep learning.

Subject Added Entry-Topical Term  

Demographics.

Subject Added Entry-Topical Term  

Benchmarks.

Subject Added Entry-Topical Term  

Demography.

Subject Added Entry-Topical Term  

Pathology.

Added Entry-Corporate Name  

Stanford University.

Host Item Entry  

Dissertations Abstracts International. 86-04B.

Electronic Location and Access  

로그인을 한후 보실 수 있는 자료입니다.

Control Number  

joongbu:658405