서브메뉴
검색
Scaling Graph Neural Networks for Sciences
Scaling Graph Neural Networks for Sciences
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202103609
- ISBN
- 9798288865367
- DDC
- 004
- 저자명
- Tripathy, Alok.
- 서명/저자
- Scaling Graph Neural Networks for Sciences
- 발행사항
- [Sl] : University of California, Berkeley, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 100 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-01, Section: B.
- 주기사항
- Advisor: Yelick, Katherine;Buluc, Aydin.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2025.
- 초록/해제
- 요약Graph Neural Networks (GNNs) have underpinned state-of-the-art results for a wide array of problems in recommender systems, molecular dynamics simulations, high-energy physics, and many other fields. However, graph datasets in these disciplines are frequently large. For instance, GNNs for high-energy physics problems contain trillions of edges. As a consequence, methods must be developed to train large-scale GNN models on massive supercomputers.However, scaling GNNs on supercomputers is difficult. GNNs are currently losers of the "hardware lottery." This means that their underlying kernels do not efficiently use the resources of a GPU-based supercomputer, unlike models like Transformers that achieve high cluster utilization. This thesis outlines methods to accelerate distributed GNN training, making GNNs practical for downstream applications in pursuit of "winning" the hardware lottery.The main ideas underlying this work are to express GNN training with sparse matrix multiplication and scale training with communication-avoiding distributed sparse matrix algorithms. First, we outline how distributed sparse-dense matrix multiplication effectively scales full-batch GNN training. We dive further into this topic by introducing efficient sparsity-aware and load balancing techniques. Second, we show how to accelerate minibatch GNN training with sparse-sparse matrix multiplication to perform batch sampling. Lastly, we show how these methods can scale GNNs that solve particle track reconstruction, an important problem in high-energy physics.
- 일반주제명
- Computer science
- 일반주제명
- Computer engineering
- 일반주제명
- Applied physics
- 일반주제명
- Information technology
- 키워드
- Graph datasets
- 키워드
- Hardware lottery
- 기타저자
- University of California, Berkeley Electrical Engineering & Computer Sciences
- 기본자료저록
- Dissertations Abstracts International. 87-01B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017357856
■00520260202103609
■006m o d
■007cr#unu||||||||
■020 ▼a9798288865367
■035 ▼a(MiAaPQ)AAI32043058
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aTripathy, Alok.
■24510▼aScaling Graph Neural Networks for Sciences
■260 ▼a[Sl]▼bUniversity of California, Berkeley▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a100 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-01, Section: B.
■500 ▼aAdvisor: Yelick, Katherine;Buluc, Aydin.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2025.
■520 ▼aGraph Neural Networks (GNNs) have underpinned state-of-the-art results for a wide array of problems in recommender systems, molecular dynamics simulations, high-energy physics, and many other fields. However, graph datasets in these disciplines are frequently large. For instance, GNNs for high-energy physics problems contain trillions of edges. As a consequence, methods must be developed to train large-scale GNN models on massive supercomputers.However, scaling GNNs on supercomputers is difficult. GNNs are currently losers of the "hardware lottery." This means that their underlying kernels do not efficiently use the resources of a GPU-based supercomputer, unlike models like Transformers that achieve high cluster utilization. This thesis outlines methods to accelerate distributed GNN training, making GNNs practical for downstream applications in pursuit of "winning" the hardware lottery.The main ideas underlying this work are to express GNN training with sparse matrix multiplication and scale training with communication-avoiding distributed sparse matrix algorithms. First, we outline how distributed sparse-dense matrix multiplication effectively scales full-batch GNN training. We dive further into this topic by introducing efficient sparsity-aware and load balancing techniques. Second, we show how to accelerate minibatch GNN training with sparse-sparse matrix multiplication to perform batch sampling. Lastly, we show how these methods can scale GNNs that solve particle track reconstruction, an important problem in high-energy physics.
■590 ▼aSchool code: 0028.
■650 4▼aComputer science
■650 4▼aComputer engineering
■650 4▼aApplied physics
■650 4▼aInformation technology
■653 ▼aGraph Neural Networks
■653 ▼aGraph datasets
■653 ▼aHigh-energy physics
■653 ▼aGPU-based supercomputer
■653 ▼aHardware lottery
■690 ▼a0984
■690 ▼a0489
■690 ▼a0464
■690 ▼a0215
■71020▼aUniversity of California, Berkeley▼bElectrical Engineering & Computer Sciences.
■7730 ▼tDissertations Abstracts International▼g87-01B.
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17357856▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
