서브메뉴
검색
Deliberate Visual-Symbolic Reasoning in a Cognitive Architecture
Deliberate Visual-Symbolic Reasoning in a Cognitive Architecture
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105220
- ISBN
- 9798291566091
- DDC
- 004
- 저자명
- Boggs, James M.
- 서명/저자
- Deliberate Visual-Symbolic Reasoning in a Cognitive Architecture
- 발행사항
- [Sl] : University of Michigan, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 125 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-03, Section: B.
- 주기사항
- Advisor: Laird, John.
- 학위논문주기
- Thesis (Ph.D.)--University of Michigan, 2025.
- 초록/해제
- 요약Current artificial intelligence systems excel at specific visual tasks but lack the ability to perform deliberate, step-by-step visual reasoning. They cannot, for example, mentally construct and manipulate an image to solve a novel problem in the way a person might imagine rearranging furniture or building with Legos. This dissertation addresses this gap by developing and evaluating a cognitive architecture for integrated, deliberate visual-symbolic reasoning.I introduce SVS 2, an extension of the Soar cognitive architecture that integrates new visual memories and processes with Soar's established symbolic reasoning capabilities. SVS 2 features a short-term Visual Working Memory (VWM) that allows an agent to construct and execute visual operation graphs, and a Visual Long-term Memory (VLTM) that stores persistent, cross-representational knowledge. This structure enables the agent to deliberately select and apply visual operations - including translation between representations, image manipulation, and learning - using its symbolic knowledge to guide the visual reasoning process and vice versa. The system supports both high-fidelity depictive (image-like) representations and more abstract descriptive representations, including those derived from neural networks.The capabilities of SVS 2 are demonstrated in two novel task domains. The Visual Character Domain establishes foundational integration by requiring an agent to perform tasks like visualizing the word "wow," rotating it 180 degrees, and recognizing the result as "mom." The more complex Image Factory domain tests multi-step visual problem-solving, tasking the agent with designing a sequence of visual transformations to construct a target product from a set of input parts. The performance of agents in these domains proves the plausibility of this "symbol-first" approach, demonstrating a functional architecture for holistic, deliberative visual-symbolic reasoning.
- 일반주제명
- Computer science
- 일반주제명
- Computer engineering
- 일반주제명
- Information technology
- 키워드
- Computer vision
- 키워드
- Neuro-symbolic
- 기타저자
- University of Michigan Computer Science & Engineering
- 기본자료저록
- Dissertations Abstracts International. 87-03B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359824
■00520260202105220
■006m o d
■007cr#unu||||||||
■020 ▼a9798291566091
■035 ▼a(MiAaPQ)AAI32271798
■035 ▼a(MiAaPQ)umichrackham006238
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aBoggs, James M.
■24510▼aDeliberate Visual-Symbolic Reasoning in a Cognitive Architecture
■260 ▼a[Sl]▼bUniversity of Michigan▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a125 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-03, Section: B.
■500 ▼aAdvisor: Laird, John.
■5021 ▼aThesis (Ph.D.)--University of Michigan, 2025.
■520 ▼aCurrent artificial intelligence systems excel at specific visual tasks but lack the ability to perform deliberate, step-by-step visual reasoning. They cannot, for example, mentally construct and manipulate an image to solve a novel problem in the way a person might imagine rearranging furniture or building with Legos. This dissertation addresses this gap by developing and evaluating a cognitive architecture for integrated, deliberate visual-symbolic reasoning.I introduce SVS 2, an extension of the Soar cognitive architecture that integrates new visual memories and processes with Soar's established symbolic reasoning capabilities. SVS 2 features a short-term Visual Working Memory (VWM) that allows an agent to construct and execute visual operation graphs, and a Visual Long-term Memory (VLTM) that stores persistent, cross-representational knowledge. This structure enables the agent to deliberately select and apply visual operations - including translation between representations, image manipulation, and learning - using its symbolic knowledge to guide the visual reasoning process and vice versa. The system supports both high-fidelity depictive (image-like) representations and more abstract descriptive representations, including those derived from neural networks.The capabilities of SVS 2 are demonstrated in two novel task domains. The Visual Character Domain establishes foundational integration by requiring an agent to perform tasks like visualizing the word "wow," rotating it 180 degrees, and recognizing the result as "mom." The more complex Image Factory domain tests multi-step visual problem-solving, tasking the agent with designing a sequence of visual transformations to construct a target product from a set of input parts. The performance of agents in these domains proves the plausibility of this "symbol-first" approach, demonstrating a functional architecture for holistic, deliberative visual-symbolic reasoning.
■590 ▼aSchool code: 0127.
■650 4▼aComputer science
■650 4▼aComputer engineering
■650 4▼aInformation technology
■653 ▼aComputer vision
■653 ▼aNeuro-symbolic
■653 ▼aCognitive architecture
■653 ▼aVisual Working Memory
■653 ▼aVisual Character Domain
■690 ▼a0984
■690 ▼a0800
■690 ▼a0489
■690 ▼a0464
■71020▼aUniversity of Michigan▼bComputer Science & Engineering.
■7730 ▼tDissertations Abstracts International▼g87-03B.
■790 ▼a0127
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359824▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
