서브메뉴
검색
Discovery of Galaxy-Scale Strong Gravitational Lenses in the Dark Energy Survey with Machine Learning
Discovery of Galaxy-Scale Strong Gravitational Lenses in the Dark Energy Survey with Machine Learning
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105131
- ISBN
- 9798291553664
- DDC
- 523
- 서명/저자
- Discovery of Galaxy-Scale Strong Gravitational Lenses in the Dark Energy Survey with Machine Learning
- 발행사항
- [Sl] : The University of Wisconsin - Madison, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 151 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-02, Section: B.
- 주기사항
- Advisor: Bechtol, Keith.
- 학위논문주기
- Thesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
- 초록/해제
- 요약Strong gravitational lensing, where a massive foreground galaxy bends and magnifies light from a more distant source, offers a unique window into the underlying cosmology of the universe and serves as a powerful tool for studying dark energy. However, identifying these rare systems in astronomical surveys remains a major challenge due to the high rate of false positives in recently applied machine learning (ML)-based search methods. In this thesis, I present two projects aimed at improving lens discovery for upcoming wide-area surveys such as LSST and Euclid. First, I develop a Vision Transformer-based ML search trained on a multi-class dataset constructed through Interactive Machine Learning to explicitly target common sources of false positives. I applied this method to over 108 images from the Dark Energy Survey (DES), with ML-selected candidates visually inspected first by citizen scientists on Zooniverse and then by strong lensing experts. This search led to the discovery of hundreds of strong lensing candidates with significantly higher precision than previous efforts. Second, I compile the outputs of three independent ML-based lens searches applied to DES to evaluate their individual performance and explore their complementarity. Using expert classifications, I show that successive ML efforts achieve improved performance, that individual models complement each other to enhance completeness, and that ensemble methods can dramatically reduce false positives. Together, these results establish scalable ML-based methodologies for strong lens discovery, laying the groundwork for efficiently processing the massive datasets expected from future surveys.
- 일반주제명
- Astrophysics
- 일반주제명
- Astronomy
- 일반주제명
- Computational physics
- 키워드
- Machine learning
- 키워드
- Cosmology
- 키워드
- False positives
- 기타저자
- The University of Wisconsin - Madison Physics
- 기본자료저록
- Dissertations Abstracts International. 87-02B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359519
■00520260202105131
■006m o d
■007cr#unu||||||||
■020 ▼a9798291553664
■035 ▼a(MiAaPQ)AAI32239235
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a523
■1001 ▼aGonzalez, Jimena.
■24510▼aDiscovery of Galaxy-Scale Strong Gravitational Lenses in the Dark Energy Survey with Machine Learning
■260 ▼a[Sl]▼bThe University of Wisconsin - Madison▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a151 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-02, Section: B.
■500 ▼aAdvisor: Bechtol, Keith.
■5021 ▼aThesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
■520 ▼aStrong gravitational lensing, where a massive foreground galaxy bends and magnifies light from a more distant source, offers a unique window into the underlying cosmology of the universe and serves as a powerful tool for studying dark energy. However, identifying these rare systems in astronomical surveys remains a major challenge due to the high rate of false positives in recently applied machine learning (ML)-based search methods. In this thesis, I present two projects aimed at improving lens discovery for upcoming wide-area surveys such as LSST and Euclid. First, I develop a Vision Transformer-based ML search trained on a multi-class dataset constructed through Interactive Machine Learning to explicitly target common sources of false positives. I applied this method to over 108 images from the Dark Energy Survey (DES), with ML-selected candidates visually inspected first by citizen scientists on Zooniverse and then by strong lensing experts. This search led to the discovery of hundreds of strong lensing candidates with significantly higher precision than previous efforts. Second, I compile the outputs of three independent ML-based lens searches applied to DES to evaluate their individual performance and explore their complementarity. Using expert classifications, I show that successive ML efforts achieve improved performance, that individual models complement each other to enhance completeness, and that ensemble methods can dramatically reduce false positives. Together, these results establish scalable ML-based methodologies for strong lens discovery, laying the groundwork for efficiently processing the massive datasets expected from future surveys.
■590 ▼aSchool code: 0262.
■650 4▼aAstrophysics
■650 4▼aAstronomy
■650 4▼aComputational physics
■653 ▼aMachine learning
■653 ▼aCosmology
■653 ▼aDark Energy Survey
■653 ▼aFalse positives
■653 ▼aVision Transformer
■690 ▼a0596
■690 ▼a0216
■690 ▼a0606
■71020▼aThe University of Wisconsin - Madison▼bPhysics.
■7730 ▼tDissertations Abstracts International▼g87-02B.
■790 ▼a0262
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359519▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
