Detail View
Neural Mechanisms of Sequence Generalization in Underspecified Environments
Citations
WEB OF SCIENCE
Citations
SCOPUS
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.advisor | 유우경 | - |
| dc.contributor.author | Taehyun Yoo | - |
| dc.date.accessioned | 2026-09-01T19:29:20Z | - |
| dc.date.available | 2026-09-01T19:29:20Z | - |
| dc.date.issued | 2026 | - |
| dc.identifier.uri | https://scholar.dgist.ac.kr/handle/20.500.11750/60717 | - |
| dc.identifier.uri | http://dgist.dcollection.net/common/orgView/200001006842 | - |
| dc.description | sequential rule generalization, functional magnetic resonance imaging (fMRI), reinforcement learning, mixture-of-experts, abstract task representation | - |
| dc.description.abstract | The ability to generalize abstract rules is fundamental to adaptive behavior in novel environments. Although previous studies have implicated frontoparietal regions in generalization, most have focused on perceptually similar stimuli or fully specified task structures, leaving unresolved how humans generalize latent sequential rules in environments supporting multiple plausible strategies. Here, we investigated the neural mechanisms underlying sequential rule generalization in unspecified environments. Participants performed a cue-based sequence learning task in which rules learned for a subset of cues could be generalized to novel cues through multiple possible cue–rule factorizations. This design created an unspecified environment in which successful performance should require the generation, evaluation, and selection of competing candidate rule structures. Behaviorally, participants with prior rule knowledge discovered the novel rules more successfully than controls. FMRI analyses revealed increased frontoparietal engagement during generalization, consistent with elevated cognitive control demands during rule inference. Outcome-dependent learning signals during feedback processing were associated with activity in the lateral orbitofrontal cortex (lOFC). In addition, representational similarity analysis demonstrated that medial orbitofrontal cortex (mOFC) representations reflected the latent structure of sequential rules over learning, selectively in participants who successfully generalized. Moreover, increases in mOFC representations of latent rule structure were associated with learning-related signals, suggesting that outcome-dependent learning contributes to the emergence of structured rule representations. Together, these findings indicate that sequential rule generalization in unspecified environments depends on coordinated interactions among cognitive control systems, outcome-dependent learning mechanisms, and representations of abstract task structure.|인간은 새로운 환경에서도 기존 지식을 활용하여 빠르게 적응할 수 있으며, 이러한 능력은 일반화의 핵심적인 특징으로 간주된다. 기존 연구들은 주로 지각적 유사성이나 명시적으로 주어진 관계 구조에 기반한 일반화에 초점을 맞추어 왔으나, 실제 환경에서는 동일한 정보로부터 여러 가능한 전략이 도출될 수 있는 불확정한 상황이 빈번하게 발생한다. 본 연구는 이러한 환경에서 인간이 어떻게 추상적 배열 규칙을 일반화하는지, 그리고 이를 뒷받침하는 신경 및 계산 메커니즘이 무엇인지를 규명하고자 하였다. 이를 위해 단서에 기반한 배열 학습 과제를 설계하였으며, 참가자들은 일부 단서에 대해서만 사전 학습된 규칙을 바탕으로 새로운 단서의 규칙을 추론해야 했다. 과제는 여러 가능한 단서-규칙 요인화를 허용하도록 구성되어 구조적으로 불확정한 환경을 형성하였다. 행동 자료는 강화학습(reinforcement learning) 및 mixture-of-experts(MoE) 모델을 통해 분석하였고, 기능적 자기공명영상(fMRI)을 이용하여 일반화 과정의 신경 활동을 측정하였다. 행동 결과, 사전 지식을 가진 참가자들은 새로운 규칙을 더 빠르게 일반화하였다. 계산 모델 분석에서는 참가자들이 피드백에 기반하여 여러 후보 전략의 가치를 평가·갱신하며 최적 전략으로 수렴하는 양상이 관찰되었다. fMRI 분석에서는 일반화 과정 동안 전두엽과 두정엽 영역의 활성 증가가 나타났으며, 이는 새로운 규칙 추론 과정에서 인지적 통제 요구가 증가함을 시사한다. 또한 선조체(putamen)와 외측 안와전두피질(lateral orbitofrontal cortex, lOFC)은 가치 및 비부호화 보상예측오차와 관련된 학습 신호를 반영하였다. 심리생리적 상호작용(PPI) 분석에서는 putamen과 복내측 전전두피질(vmPFC) 간 기능적 연결성이 일반화 과정에서 증가하였다. 표상유사도분석(RSA) 결과, 내측 안와전두피질(mOFC)은 학습이 진행됨에 따라 올바른 전략의 잠재적 구조를 점진적으로 표상하였으며, 이러한 구조화된 표상은 일반화에 성공한 참가자들에게서 선택적으로 나타났다. 또한 mOFC 표상의 증가는 비부호화 보상예측오차의 감소와 음의 상관관계를 보여, 결과 기반 학습 신호가 구조화된 인지 지도 형성에 기여함을 시사하였다. 종합하면, 본 연구는 추상적 배열 규칙 일반화가 인지적 통제, 가치 기반 학습, 그리고 안와전두피질 기반의 구조화된 과제 표상의 상호작용을 통해 이루어짐을 제안한다. 이러한 결과는 인간이 불확정 환경에서 사전 지식을 재구성하여 새로운 내부 과제 모델을 형성하고 유연하게 일반화할 수 있음을 보여주며, 일반화의 신경계산적 메커니즘에 대한 새로운 이해를 제공한다. |
- |
| dc.description.tableofcontents | Ⅰ. Introduction 1 1.1 Preface 1 1.2 Research Background 2 1.2.1 Generalization as adaptive cognition 2 1.2.2 Abstract task representations and cognitive maps 2 1.3 Limitations of Previous Research 2 1.3.1 Emphasis on perceptual similarity 2 1.3.2 Limited consideration of unspecified environments and competing strategies 3 1.4 The Present Study 3 1.4.1 Sequential structure as a domain for abstract generalization 3 1.4.2 Factorization and the generation of candidate strategies 3 1.4.3 Research objectives and hypotheses 4 1.5 Overview of the Thesis 5 ⅠI. Methods 7 2.1 Participants 7 2.2 Experimental Design 8 2.3 Behavioral Data Analyses 9 2.4 Computational Modeling Analyses 10 2.4.1 Computational model 1: Reinforcement Learning model 10 2.4.2 Computational model 2: Mixture-of-Experts model 14 2.5 fMRI Data Acquisition and Preprocessing 18 2.6 fMRI Data Analyses 18 2.6.1 Univariate Analysis 18 2.6.2 Parametric Modulation Analysis 20 2.6.3 Psychophysiological Interaction Analysis 21 2.6.4 Representational Similarity Analysis 22 III. Results 25 3.1 Participants generalize trained rules to acquire novel sequential rules 25 3.2 Reinforcement learning models capture individual differences in valuation and strategy use 29 3.3 Mixture-of-experts models capture arbitration between competing strategies 31 3.4 Neural sensitivity to hierarchical sequence structure during trained sequence processing 34 3.5 Frontoparietal control regions support active generalization of novel sequential rules 36 3.6 Attention-related network recruitment differs between generalizers and non-generalizers 47 3.7 Striatal and orbitofrontal learning signals support rule inference and generalization 49 3.8 Learning-related representational organization emerges in the medial OFC during rule generalization 53 IV. Discussion 58 4.1 Significance of the Present Study 58 4.2 The frontoparietal regions as action controller during rule generalization 59 4.3 The valuation circuitry as value arbiter of competing generalization strategies 60 4.4 Frontostriatal coupling as integrator of value and strategy evaluation 60 4.5 The cognitive map system as strategy planner for abstract task generalization 62 V. Conclusion 64 References 65 요약문 68 |
- |
| dc.format.extent | 68 | - |
| dc.language | eng | - |
| dc.publisher | DGIST | - |
| dc.title | Neural Mechanisms of Sequence Generalization in Underspecified Environments | - |
| dc.title.alternative | 불확정 환경에서의 추상적 배열 규칙 일반화와 관련된 신경 기저 | - |
| dc.type | Thesis | - |
| dc.identifier.doi | 10.22677/THESIS.200001006842 | - |
| dc.description.degree | Doctor | - |
| dc.contributor.department | Department of Brain Sciences | - |
| dc.contributor.coadvisor | Hyeon-Ae Jeon | - |
| dc.date.awarded | 2026-08-01 | - |
| dc.publisher.location | Daegu | - |
| dc.description.database | dCollection | - |
| dc.citation | XT.BD 유88 202608 | - |
| dc.date.accepted | 2026-07-21 | - |
| dc.contributor.alternativeDepartment | 뇌과학과 | - |
| dc.subject.keyword | sequential rule generalization, functional magnetic resonance imaging (fMRI), reinforcement learning, mixture-of-experts, abstract task representation | - |
| dc.contributor.affiliatedAuthor | Taehyun Yoo | - |
| dc.contributor.affiliatedAuthor | Wookyung Yu | - |
| dc.contributor.affiliatedAuthor | Hyeon-Ae Jeon | - |
| dc.contributor.alternativeName | 유태현 | - |
| dc.contributor.alternativeName | Wookyung Yu | - |
| dc.contributor.alternativeName | 전현애 | - |
| dc.rights.embargoReleaseDate | 2029-08-31 | - |
File Downloads
- There are no files associated with this item.
공유
Total Views & Downloads
???jsp.display-item.statistics.view???: , ???jsp.display-item.statistics.download???:
