서브메뉴
검색
Exploring Biologically-Inspired Models for Multifaceted Learning in the Brain
Exploring Biologically-Inspired Models for Multifaceted Learning in the Brain
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211151957
- ISBN
- 9798383201374
- DDC
- 153
- 저자명
- Cheng, Huzi.
- 서명/저자
- Exploring Biologically-Inspired Models for Multifaceted Learning in the Brain
- 발행사항
- [Sl] : Indiana University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 122 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-01, Section: B.
- 주기사항
- Advisor: Brown, Joshua W.
- 학위논문주기
- Thesis (Ph.D.)--Indiana University, 2024.
- 초록/해제
- 요약Unraveling the computational foundations of learning is one of the paramount quests in neuroscience. This thesis employs a computational approach to investigate this question through three distinct projects, spanning from single-cell level to cross-brain-region mechanisms. The first project proposes a viable alternative theory to Feedback Alignment (Lillicrap et al., 2014), a mechanism suggested as a replacement for backpropagation (Rumelhart et al., 1986) in the biological brain for learning across different layers of neurons. We explore the validity of this theory and investigate novel solutions derived from it, in addition to Feedback Alignment. The second project develops a model, R2N2, for sequence learning in recurrent neural networks. The model has stronger performance when compared with other biologically plausible sequence learning algorithms in benchmark tests and shows potential in modeling animal behaviors in a T-maze navigation task. While partly building on the results of the first project, the main aim here is to understand how the brain processes temporal sequences.The final project extends to the systemic level, devising a model, deepGOLSA, for goal-directed learning that utilizes neural representations and corresponding subgoal decompositions. The resulting solution is versatile and can be applied to tasks of arbitrary complexity. When integrated with reinforcement learning algorithms, it accelerates their performance in various discrete and continuous space tasks. When applied in isolation, it outperforms all benchmark algorithms in certain tasks. Furthermore, we used this model to simulate and analyze human behavior and brain data in a treasure hunting cognitive task. The findings offer new insights into the role of several brain regions like vmPFC in goal-directed behaviors.
- 일반주제명
- Cognitive psychology
- 일반주제명
- Neurosciences
- 일반주제명
- Systematic biology
- 키워드
- Brain
- 기타저자
- Indiana University Psychological & Brain Sciences
- 기본자료저록
- Dissertations Abstracts International. 86-01B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017162311
■00520250211151957
■006m o d
■007cr#unu||||||||
■020 ▼a9798383201374
■035 ▼a(MiAaPQ)AAI31329305
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a153
■1001 ▼aCheng, Huzi.
■24510▼aExploring Biologically-Inspired Models for Multifaceted Learning in the Brain
■260 ▼a[Sl]▼bIndiana University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a122 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-01, Section: B.
■500 ▼aAdvisor: Brown, Joshua W.
■5021 ▼aThesis (Ph.D.)--Indiana University, 2024.
■520 ▼aUnraveling the computational foundations of learning is one of the paramount quests in neuroscience. This thesis employs a computational approach to investigate this question through three distinct projects, spanning from single-cell level to cross-brain-region mechanisms. The first project proposes a viable alternative theory to Feedback Alignment (Lillicrap et al., 2014), a mechanism suggested as a replacement for backpropagation (Rumelhart et al., 1986) in the biological brain for learning across different layers of neurons. We explore the validity of this theory and investigate novel solutions derived from it, in addition to Feedback Alignment. The second project develops a model, R2N2, for sequence learning in recurrent neural networks. The model has stronger performance when compared with other biologically plausible sequence learning algorithms in benchmark tests and shows potential in modeling animal behaviors in a T-maze navigation task. While partly building on the results of the first project, the main aim here is to understand how the brain processes temporal sequences.The final project extends to the systemic level, devising a model, deepGOLSA, for goal-directed learning that utilizes neural representations and corresponding subgoal decompositions. The resulting solution is versatile and can be applied to tasks of arbitrary complexity. When integrated with reinforcement learning algorithms, it accelerates their performance in various discrete and continuous space tasks. When applied in isolation, it outperforms all benchmark algorithms in certain tasks. Furthermore, we used this model to simulate and analyze human behavior and brain data in a treasure hunting cognitive task. The findings offer new insights into the role of several brain regions like vmPFC in goal-directed behaviors.
■590 ▼aSchool code: 0093.
■650 4▼aCognitive psychology
■650 4▼aNeurosciences
■650 4▼aSystematic biology
■653 ▼aBrain
■653 ▼aFeedback alignment algorithm
■653 ▼aInformation augmentation
■653 ▼aPlausible learning algorithms
■653 ▼aHuman brain activity
■690 ▼a0633
■690 ▼a0317
■690 ▼a0800
■690 ▼a0423
■71020▼aIndiana University▼bPsychological & Brain Sciences.
■7730 ▼tDissertations Abstracts International▼g86-01B.
■790 ▼a0093
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17162311▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


