서브메뉴
검색
Bridging the Gap Between Automated Logical Reasoning and Machine Learning.
Bridging the Gap Between Automated Logical Reasoning and Machine Learning.
상세정보
- 자료유형
- 학위논문
- Control Number
- 0017162968
- International Standard Book Number
- 9798384341765
- Dewey Decimal Classification Number
- 515
- Main Entry-Personal Name
- Wu, Haoze.
- Publication, Distribution, etc. (Imprint
- [S.l.] : Stanford University., 2024
- Publication, Distribution, etc. (Imprint
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- Physical Description
- 164 p.
- General Note
- Source: Dissertations Abstracts International, Volume: 86-03, Section: A.
- General Note
- Advisor: Barrett, Clark.
- Dissertation Note
- Thesis (Ph.D.)--Stanford University, 2024.
- Summary, Etc.
- 요약Human rationality comprises two facets: deductive reasoning -- deriving conclusions from premises, and inductive reasoning -- inferring patterns from observations. These two forms of thinking are reflected in the ways we automate tasks with computers. The former plays a crucial role in the field of formal methods, which aims to ensure that computer systems work exactly as intended, in domains ranging from chip design to aviation. The latter drives the field of machine learning, which has revolutionized our ability to handle complex and uncertain inputs, such as images and natural languages. While both approaches have enabled tremendous technological progress, each faces substantial challenges: formal methods struggle with scalability and abstract high-level reasoning, while the unstable and opaque nature of deep learning models impedes their actual deployment in high-stake applications. One promising avenue towards addressing these challenges, and creating safer and smarter computer systems, is to bridge the gap between these two fields: just as human intelligence is a symbiosis of deduction and induction, I dream about a future where computer systems seamlessly integrate logical reasoning and machine learning to automate complex tasks in a scalable, reliable, explainable, and safe manner.I think my tendency to dream about this future comes from my background in philosophy during my undergraduate studies. Through a combination of serendipity, and conscious and sub-conscious choices, I have been able to explore this direction during my PhD training. Although the dream is not fully realized, I feel fortunate to have put together this disseration that may contribute to that future. The dissertation is divided into two parts, each guided by a key question:1. Can automated reasoning be used to prove properties about deep-learning-enabled systems?2. Can machine learning enhance the scalability of automated reasoning tools?Part I discusses a set of automated reasoning techniques for formally analyzing artificial neural networks, the cornerstone of modern machine learning. These techniques lie at the core of version 2.0 of the Marabou framework, a popular formal analyzer of neural networks. Part II outlines two methodologies for developing push-button machine-learning-driven automated reasoning tools, with applications in hardware and program verification.
- Subject Added Entry-Topical Term
- Convex analysis.
- Subject Added Entry-Topical Term
- Distance learning.
- Subject Added Entry-Topical Term
- Large language models.
- Subject Added Entry-Topical Term
- Cognition & reasoning.
- Subject Added Entry-Topical Term
- Neural networks.
- Subject Added Entry-Topical Term
- Cognitive psychology.
- Subject Added Entry-Topical Term
- Educational technology.
- Subject Added Entry-Topical Term
- Mathematics.
- Added Entry-Corporate Name
- Stanford University.
- Host Item Entry
- Dissertations Abstracts International. 86-03A.
- Electronic Location and Access
- 로그인을 한후 보실 수 있는 자료입니다.
- Control Number
- joongbu:655427
MARC
008250224s2024 us ||||||||||||||c||eng d■001000017162968
■00520250211152118
■006m o d
■007cr#unu||||||||
■020 ▼a9798384341765
■035 ▼a(MiAaPQ)AAI31460332
■035 ▼a(MiAaPQ)Stanfordpw103gj9221
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a515
■1001 ▼aWu, Haoze.
■24510▼aBridging the Gap Between Automated Logical Reasoning and Machine Learning.
■260 ▼a[S.l.]▼bStanford University. ▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a164 p.
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-03, Section: A.
■500 ▼aAdvisor: Barrett, Clark.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2024.
■520 ▼aHuman rationality comprises two facets: deductive reasoning -- deriving conclusions from premises, and inductive reasoning -- inferring patterns from observations. These two forms of thinking are reflected in the ways we automate tasks with computers. The former plays a crucial role in the field of formal methods, which aims to ensure that computer systems work exactly as intended, in domains ranging from chip design to aviation. The latter drives the field of machine learning, which has revolutionized our ability to handle complex and uncertain inputs, such as images and natural languages. While both approaches have enabled tremendous technological progress, each faces substantial challenges: formal methods struggle with scalability and abstract high-level reasoning, while the unstable and opaque nature of deep learning models impedes their actual deployment in high-stake applications. One promising avenue towards addressing these challenges, and creating safer and smarter computer systems, is to bridge the gap between these two fields: just as human intelligence is a symbiosis of deduction and induction, I dream about a future where computer systems seamlessly integrate logical reasoning and machine learning to automate complex tasks in a scalable, reliable, explainable, and safe manner.I think my tendency to dream about this future comes from my background in philosophy during my undergraduate studies. Through a combination of serendipity, and conscious and sub-conscious choices, I have been able to explore this direction during my PhD training. Although the dream is not fully realized, I feel fortunate to have put together this disseration that may contribute to that future. The dissertation is divided into two parts, each guided by a key question:1. Can automated reasoning be used to prove properties about deep-learning-enabled systems?2. Can machine learning enhance the scalability of automated reasoning tools?Part I discusses a set of automated reasoning techniques for formally analyzing artificial neural networks, the cornerstone of modern machine learning. These techniques lie at the core of version 2.0 of the Marabou framework, a popular formal analyzer of neural networks. Part II outlines two methodologies for developing push-button machine-learning-driven automated reasoning tools, with applications in hardware and program verification.
■590 ▼aSchool code: 0212.
■650 4▼aConvex analysis.
■650 4▼aDistance learning.
■650 4▼aLarge language models.
■650 4▼aCognition & reasoning.
■650 4▼aNeural networks.
■650 4▼aCognitive psychology.
■650 4▼aEducational technology.
■650 4▼aMathematics.
■690 ▼a0800
■690 ▼a0633
■690 ▼a0710
■690 ▼a0405
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g86-03A.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17162968▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
