서브메뉴
검색
Through the Recurrent Neural Network Looking Glass: Structure-Function Relationships in Cortical Circuits for Predictive Coding
Through the Recurrent Neural Network Looking Glass: Structure-Function Relationships in Cortical Circuits for Predictive Coding
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105531
- ISBN
- 9798263344221
- DDC
- 006
- 서명/저자
- Through the Recurrent Neural Network Looking Glass: Structure-Function Relationships in Cortical Circuits for Predictive Coding
- 발행사항
- [Sl] : Georgia Institute of Technology, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 190 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Choi, Hannah;Rozell, Christopher.
- 학위논문주기
- Thesis (Ph.D.)--Georgia Institute of Technology, 2025.
- 초록/해제
- 요약The elucidation of structure-function relationships in neural circuits forms the cornerstone of theoretical and computational neuroscience. It lays the foundation for our understanding of information representation and processing in the brain, influencing normative theories of neural dynamics, cognition, and behaviour. It also sheds light on the mechanistic properties of these circuits, thereby guiding research in systems neuroscience, artificial intelligence, and machine learning. However, despite decades of neuroscientific inquiry and technological advancements, our grasp on how neural architecture shapes its computations remains incomplete. This limitation is particularly evident in our understanding of the cortex, where its intricate laminar structure is arranged into the recurring neuroanatomical motif known as the canonical cortical microcircuit. While we have extensive knowledge of the structural scaffolding itself, we lack a comprehensive understanding of how this architecture supports its sophisticated information processing. Specifically, the precise mechanisms by which the canonical microcircuit and its inter-areal connections facilitate predictive coding, a theoretical framework positing that the cortex implements hierarchical, predictive computations for efficient perception and inference, remains elusive.Unfortunately, in-vivo efforts to experimentally test the underlying mechanisms of predictive coding still face enormous challenges. Furthermore, even as recent experimental advances have begun to yield interesting multi-regional neuronal recordings, our ability to reliably interpret these datasets at scale remains limited. In-silico approaches, therefore, present an encouraging avenue to address these shortcomings. Specifically, recurrent neural networks (RNNs) offer a powerful framework to investigate hypotheses about hierarchical predictive coding that are currently challenging to test in-vivo, owing to their ability to effectively exploit temporal dependencies typical of natural stimuli, learn distributed representations, and model complex, multi-regional neuronal dynamics.Hence, this thesis employs RNNs as models of cortical circuits, illustrating how they can be leveraged to study the inherent architectural biases of the canonical cortical microcircuit and their functional implications, in light of the predictive coding hypothesis. Specifically, it i) Establishes a computational framework using RNNs and representational analyses to study the effects of biologically-motivated inter-areal laminar connections on the computational roles of different neuronal populations in the microcircuit across hierarchically-related areas, ii) Demonstrates how feedback connections combined with time-delays provide an inductive bias for differentiating expected and unexpected inputs, and iii) Studies the effects of including a predictive coding-based training strategy on the representational geometry of neuronal populations receiving inter-areal feedback. Furthermore, this dissertation also develops tools that address limitations concerning the anatomical fidelity of conventional RNN architecture and training, thereby improving their overall biological plausibility and potential for modelling real neural circuits. More concretely, the key contributions include mathematically-grounded methods with performance guarantees for i) Training RNNs that respect Dale's law, and ii) Pruning RNN weights to enforce highly sparse, structured connectivity patterns amongst different neuronal populations. Finally, applying these methods to analyze multi-regional interactions in 2-photon calcium imaging data from mice performing a predictive visual task corroborates experimental evidence in support of the predictive coding hypothesis, showing the utility of the proposed methods and overall approach.The contributions of this thesis therefore provide a robust, computational framework for modelling and interpreting complex, multi-regional neuronal data, ultimately paving the way for deeper insights into how the structural properties of cortical circuits support sophisticated information processing and intelligent behavior.
- 일반주제명
- Deep learning
- 일반주제명
- Back propagation
- 일반주제명
- Neural networks
- 일반주제명
- Neurosciences
- 일반주제명
- Information processing
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360467
■00520260202105531
■006m o d
■007cr#unu||||||||
■020 ▼a9798263344221
■035 ▼a(MiAaPQ)AAI32309842
■035 ▼a(MiAaPQ)GeorgiaTech77843
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a006
■1001 ▼aBalwani, Aishwarya.
■24510▼aThrough the Recurrent Neural Network Looking Glass: Structure-Function Relationships in Cortical Circuits for Predictive Coding
■260 ▼a[Sl]▼bGeorgia Institute of Technology▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a190 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Choi, Hannah;Rozell, Christopher.
■5021 ▼aThesis (Ph.D.)--Georgia Institute of Technology, 2025.
■520 ▼aThe elucidation of structure-function relationships in neural circuits forms the cornerstone of theoretical and computational neuroscience. It lays the foundation for our understanding of information representation and processing in the brain, influencing normative theories of neural dynamics, cognition, and behaviour. It also sheds light on the mechanistic properties of these circuits, thereby guiding research in systems neuroscience, artificial intelligence, and machine learning. However, despite decades of neuroscientific inquiry and technological advancements, our grasp on how neural architecture shapes its computations remains incomplete. This limitation is particularly evident in our understanding of the cortex, where its intricate laminar structure is arranged into the recurring neuroanatomical motif known as the canonical cortical microcircuit. While we have extensive knowledge of the structural scaffolding itself, we lack a comprehensive understanding of how this architecture supports its sophisticated information processing. Specifically, the precise mechanisms by which the canonical microcircuit and its inter-areal connections facilitate predictive coding, a theoretical framework positing that the cortex implements hierarchical, predictive computations for efficient perception and inference, remains elusive.Unfortunately, in-vivo efforts to experimentally test the underlying mechanisms of predictive coding still face enormous challenges. Furthermore, even as recent experimental advances have begun to yield interesting multi-regional neuronal recordings, our ability to reliably interpret these datasets at scale remains limited. In-silico approaches, therefore, present an encouraging avenue to address these shortcomings. Specifically, recurrent neural networks (RNNs) offer a powerful framework to investigate hypotheses about hierarchical predictive coding that are currently challenging to test in-vivo, owing to their ability to effectively exploit temporal dependencies typical of natural stimuli, learn distributed representations, and model complex, multi-regional neuronal dynamics.Hence, this thesis employs RNNs as models of cortical circuits, illustrating how they can be leveraged to study the inherent architectural biases of the canonical cortical microcircuit and their functional implications, in light of the predictive coding hypothesis. Specifically, it i) Establishes a computational framework using RNNs and representational analyses to study the effects of biologically-motivated inter-areal laminar connections on the computational roles of different neuronal populations in the microcircuit across hierarchically-related areas, ii) Demonstrates how feedback connections combined with time-delays provide an inductive bias for differentiating expected and unexpected inputs, and iii) Studies the effects of including a predictive coding-based training strategy on the representational geometry of neuronal populations receiving inter-areal feedback. Furthermore, this dissertation also develops tools that address limitations concerning the anatomical fidelity of conventional RNN architecture and training, thereby improving their overall biological plausibility and potential for modelling real neural circuits. More concretely, the key contributions include mathematically-grounded methods with performance guarantees for i) Training RNNs that respect Dale's law, and ii) Pruning RNN weights to enforce highly sparse, structured connectivity patterns amongst different neuronal populations. Finally, applying these methods to analyze multi-regional interactions in 2-photon calcium imaging data from mice performing a predictive visual task corroborates experimental evidence in support of the predictive coding hypothesis, showing the utility of the proposed methods and overall approach.The contributions of this thesis therefore provide a robust, computational framework for modelling and interpreting complex, multi-regional neuronal data, ultimately paving the way for deeper insights into how the structural properties of cortical circuits support sophisticated information processing and intelligent behavior.
■590 ▼aSchool code: 0078.
■650 4▼aDeep learning
■650 4▼aBack propagation
■650 4▼aNeural networks
■650 4▼aNeurosciences
■650 4▼aInformation processing
■690 ▼a0800
■690 ▼a0317
■71020▼aGeorgia Institute of Technology.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0078
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360467▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


