서브메뉴
검색
Deep Learning Frameworks for Accelerating Catalyst Simulations
Deep Learning Frameworks for Accelerating Catalyst Simulations
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211153146
- ISBN
- 9798896070160
- DDC
- 660
- 저자명
- Kolluru, Adeesh.
- 서명/저자
- Deep Learning Frameworks for Accelerating Catalyst Simulations
- 발행사항
- [Sl] : Carnegie Mellon University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 136 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-04, Section: B.
- 주기사항
- Advisor: Kitchin, John R.
- 학위논문주기
- Thesis (Ph.D.)--Carnegie Mellon University, 2024.
- 초록/해제
- 요약Machine learning (ML) methods have the potential to solve the longstanding challenge of discovering catalysts that can convert renewable energy to fuel, discovering molecules with desired properties, and discovering drugs as they are up to O(105) times faster than the conventional quantum mechanical (QM) simulators like Density Functional Theory (DFT). Graph Neural Networks (GNNs) have emerged as promising approximators of DFT but face challenges with generalization. Moreover, for a few simulations replacing DFT with GNNs is not sufficient to perform large-scale screening and requires the development of fundamentally novel approaches to accelerate simulations.This thesis explores advanced machine learning (ML) methodologies and frameworks for accelerating materials discovery, with a primary focus on heterogeneous catalysis. The research addresses challenges in large-scale ML applications within catalysis, examining out-of-distribution generalization, metrics, and methodological approaches. A novel transfer learning framework is proposed to improve the generalization of the model across atomic domains. While generalizable ML potentials significantly expedite catalyst simulation by substituting density functional theory (DFT), the thesis demonstrates that novel ML frameworks can further accelerate simulations. To this end, a GNN model is developed to efficiently learn and directly predict the optimal adsorbate-catalyst configurations. The research culminates in the development of a diffusion framework designed to accelerate adsorbate-catalyst optimization processes.
- 일반주제명
- Chemical engineering
- 일반주제명
- Computer engineering
- 일반주제명
- Energy
- 키워드
- Diffusion models
- 키워드
- Machine learning
- 기타저자
- Carnegie Mellon University Chemical Engineering
- 기본자료저록
- Dissertations Abstracts International. 86-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017165187
■00520250211153146
■006m o d
■007cr#unu||||||||
■020 ▼a9798896070160
■035 ▼a(MiAaPQ)AAI31565469
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a660
■1001 ▼aKolluru, Adeesh.▼0(orcid)0000-0001-8125-6881
■24510▼aDeep Learning Frameworks for Accelerating Catalyst Simulations
■260 ▼a[Sl]▼bCarnegie Mellon University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a136 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-04, Section: B.
■500 ▼aAdvisor: Kitchin, John R.
■5021 ▼aThesis (Ph.D.)--Carnegie Mellon University, 2024.
■520 ▼aMachine learning (ML) methods have the potential to solve the longstanding challenge of discovering catalysts that can convert renewable energy to fuel, discovering molecules with desired properties, and discovering drugs as they are up to O(105) times faster than the conventional quantum mechanical (QM) simulators like Density Functional Theory (DFT). Graph Neural Networks (GNNs) have emerged as promising approximators of DFT but face challenges with generalization. Moreover, for a few simulations replacing DFT with GNNs is not sufficient to perform large-scale screening and requires the development of fundamentally novel approaches to accelerate simulations.This thesis explores advanced machine learning (ML) methodologies and frameworks for accelerating materials discovery, with a primary focus on heterogeneous catalysis. The research addresses challenges in large-scale ML applications within catalysis, examining out-of-distribution generalization, metrics, and methodological approaches. A novel transfer learning framework is proposed to improve the generalization of the model across atomic domains. While generalizable ML potentials significantly expedite catalyst simulation by substituting density functional theory (DFT), the thesis demonstrates that novel ML frameworks can further accelerate simulations. To this end, a GNN model is developed to efficiently learn and directly predict the optimal adsorbate-catalyst configurations. The research culminates in the development of a diffusion framework designed to accelerate adsorbate-catalyst optimization processes.
■590 ▼aSchool code: 0041.
■650 4▼aChemical engineering
■650 4▼aComputer engineering
■650 4▼aEnergy
■653 ▼aComputational catalysis
■653 ▼aDiffusion models
■653 ▼aGraph Neural Networks
■653 ▼aMachine learning
■653 ▼aTransfer learning
■690 ▼a0542
■690 ▼a0464
■690 ▼a0791
■71020▼aCarnegie Mellon University▼bChemical Engineering.
■7730 ▼tDissertations Abstracts International▼g86-04B.
■790 ▼a0041
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17165187▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
