Yan Jiao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yan Jiao has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Limei Peng
Kyungpook National University
10 shared publications Pin-Han Ho
5 shared publications- Pin-Han Ho
University of Electronic Science and Technology of China
5 shared publications - Yiming Miao
University of Electronic Science and Technology of China
4 shared publications - Shih Yu Chang
Department of Applied Data Science, San Jose State University, San Jose, CA 95192, USA
2 shared publications - Jingran Xu1 shared publication
Latest Publications
- 24 October 2026
- Published by Elsevier in Information Sciences
- Vol. 765
Sensor-conditioned representation learning for structured sensing via scene-relevant observation quotients
- 24 October 2026
- Published by Elsevier in Information Sciences
- Vol. 765
Sensor-conditioned representation learning for structured sensing via scene-relevant observation quotients
- Open Access
- 16 September 2026
- Published by Elsevier in SSRN Electronic Journal
SemanticBelief: Persistent Semantic Belief for Dynamic 3D Recognition under Partial Observation
Dynamic 3D pattern recognition under partial observation is fundamentally ambiguous because missing or degraded evidence does not determine whether an entity has disappeared, remains temporarily unobserved, or still supports a reliable semantic interpretation. Existing 3D recognition, temporal association, tracking, and uncertainty methods address complementary aspects ofthis problem, but typically do not expose observation support, entity persistence, semantic reliability, and residual uncertainty jointly as an explicit persistent recognition state. We introduce SemanticBelief, an entity-centric persistent semantic belief representation that maintains these four quantities as separately addressable coordinates of a recursively updated state. This organization allows observability, persistence, semantic interpretation, and uncertainty to evolve differently as evidence becomes missing, degraded, conflicting, or available again.We evaluate SemanticBelief in three complementary regimes. SemanticKITTI provides real-LiDAR grounding under controlled observation degradation and recovery. The Waymo Open Dataset evaluates persistent semantic-state evolution under real dynamic trajectories, where SemanticBelief exhibits modestly highercontext-level semantic correctness than frame-local prediction and causal exponential moving average while maintaining comparable long-horizon stability. Controlled Dynamic 3D (CD3D) experiments further evaluate evidence-dependent belief evolution, temporary absence and permanent removal, structured readout, and coordinate-level behavior using evaluator-only oracleannotations. Together, the results support persistent semantic belief as an explicit recognition-state representation for dynamic 3D perception under incomplete and changing evidence.
- 16 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSemanticBelief: Persistent Semantic Belief for Dynamic 3D Recognition under Partial Observation
Dynamic 3D pattern recognition under partial observation is fundamentally ambiguous because missing or degraded evidence does not determine whether an entity has disappeared, remains temporarily unobserved, or still supports a reliable semantic interpretation. Existing 3D recognition, temporal association, tracking, and uncertainty methods address complementary aspects ofthis problem, but typically do not expose observation support, entity persistence, semantic reliability, and residual uncertainty jointly as an explicit persistent recognition state. We introduce SemanticBelief, an entity-centric persistent semantic belief representation that maintains these four quantities as separately addressable coordinates of a recursively updated state. This organization allows observability, persistence, semantic interpretation, and uncertainty to evolve differently as evidence becomes missing, degraded, conflicting, or available again.We evaluate SemanticBelief in three complementary regimes. SemanticKITTI provides real-LiDAR grounding under controlled observation degradation and recovery. The Waymo Open Dataset evaluates persistent semantic-state evolution under real dynamic trajectories, where SemanticBelief exhibits modestly highercontext-level semantic correctness than frame-local prediction and causal exponential moving average while maintaining comparable long-horizon stability. Controlled Dynamic 3D (CD3D) experiments further evaluate evidence-dependent belief evolution, temporary absence and permanent removal, structured readout, and coordinate-level behavior using evaluator-only oracleannotations. Together, the results support persistent semantic belief as an explicit recognition-state representation for dynamic 3D perception under incomplete and changing evidence.
- Open Access
- 16 September 2026
- Published by Elsevier in SSRN Electronic Journal
Semantic Memory Regulation: A Tensor-Based Framework for Resource-Constrained Continual Learning
Continually adaptive intelligent systems must acquire new knowledge while preserving prior capabilities under finite memory resources. This creates a fundamental regulation problem: insufficient semantic sharing leads to redundant storage, whereas excessive sharing can collapse representational diversity and accelerate forgetting. We formulate continual-learning memory as an evolving bounded semantic state and develop Semantic Memory Tensors (SMT) to regulate it. SMT represents accumulated knowledge as a low-rank tensor over experience context, semantic concept, and feature dimensions. Semantic routing dynamically inserts, merges, and prunes reusable factors, while latent semantic replay reconstructs historical representations without accessing raw data. Theoretically, we characterize tensorized semantic memory capacity and derive a forgetting decomposition that separates semantic compression, capacity-constrained replay diversity, and finite replay sampling effects. We further identify a semantic sharing--collapse tradeoff, where excessive factor merging reduces replay diversity and semantic discriminability. This tradeoff reveals an optimal operating regime for memory regulation. Experiments on MNIST, CIFAR-10, and Tiny ImageNet demonstrate that SMT achieves competitive continual-learning performance with substantial memory reduction. Rank, overlap, scalability, and semantic-reuse studies empirically support the predicted theoretical behaviors. These results establish tensorized semantic memory as a promising mechanism for regulating reusable knowledge under finite-memory continual learning.
- 16 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSemantic Memory Regulation: A Tensor-Based Framework for Resource-Constrained Continual Learning
Continually adaptive intelligent systems must acquire new knowledge while preserving prior capabilities under finite memory resources. This creates a fundamental regulation problem: insufficient semantic sharing leads to redundant storage, whereas excessive sharing can collapse representational diversity and accelerate forgetting. We formulate continual-learning memory as an evolving bounded semantic state and develop Semantic Memory Tensors (SMT) to regulate it. SMT represents accumulated knowledge as a low-rank tensor over experience context, semantic concept, and feature dimensions. Semantic routing dynamically inserts, merges, and prunes reusable factors, while latent semantic replay reconstructs historical representations without accessing raw data. Theoretically, we characterize tensorized semantic memory capacity and derive a forgetting decomposition that separates semantic compression, capacity-constrained replay diversity, and finite replay sampling effects. We further identify a semantic sharing--collapse tradeoff, where excessive factor merging reduces replay diversity and semantic discriminability. This tradeoff reveals an optimal operating regime for memory regulation. Experiments on MNIST, CIFAR-10, and Tiny ImageNet demonstrate that SMT achieves competitive continual-learning performance with substantial memory reduction. Rank, overlap, scalability, and semantic-reuse studies empirically support the predicted theoretical behaviors. These results establish tensorized semantic memory as a promising mechanism for regulating reusable knowledge under finite-memory continual learning.
- Open Access
- 28 August 2026
- Published by Elsevier in SSRN Electronic Journal
Persistent Memory is Not Valid Memory: Toward Self-Maintaining Artificial Intelligence
Modern AI systems increasingly depend on persistent representations-world models, retrieved facts, and evolving knowledge structures-to support reasoning, planning, and action over time. Yet persistence does not guarantee that stored information remains justified when observations change, documents are updated, or task requirements shift. We argue that a central unresolved problem in AI memory is not merely storage or retrieval, but maintenance: determining which stored distinctions remain epistemically supportable under current evidence and task demands, and which of those supportable distinctions can still be materialized under finite compute, memory, latency, and energy budgets. These are distinct failure modes: epistemic loss calls for reobservation or revalidation, whereas resource-induced loss calls for compression, deferral, or reallocation. We advocate an explicit epistemic maintenance layer that continuously tracks both the supportable boundary and the resource-feasible maintained boundary of persistent memory. We introduce boundary-conditioned epistemic memory (BCEM) as one architectural realization of this principle, and describe memory homeostasis as the closed-loop control mechanism that keeps memory aligned with changing evidence, tasks, and resources. This perspective reframes AI memory from passive storage to active maintenance of what the system remains justified-and able-to know.
- 28 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPersistent Memory is Not Valid Memory: Toward Self-Maintaining Artificial Intelligence
Modern AI systems increasingly depend on persistent representations-world models, retrieved facts, and evolving knowledge structures-to support reasoning, planning, and action over time. Yet persistence does not guarantee that stored information remains justified when observations change, documents are updated, or task requirements shift. We argue that a central unresolved problem in AI memory is not merely storage or retrieval, but maintenance: determining which stored distinctions remain epistemically supportable under current evidence and task demands, and which of those supportable distinctions can still be materialized under finite compute, memory, latency, and energy budgets. These are distinct failure modes: epistemic loss calls for reobservation or revalidation, whereas resource-induced loss calls for compression, deferral, or reallocation. We advocate an explicit epistemic maintenance layer that continuously tracks both the supportable boundary and the resource-feasible maintained boundary of persistent memory. We introduce boundary-conditioned epistemic memory (BCEM) as one architectural realization of this principle, and describe memory homeostasis as the closed-loop control mechanism that keeps memory aligned with changing evidence, tasks, and resources. This perspective reframes AI memory from passive storage to active maintenance of what the system remains justified-and able-to know.
- Open Access
- 28 August 2026
- Published by Elsevier in SSRN Electronic Journal
Boundary-Conditioned Epistemic Memory: A Substrate for Self-Maintaining AI
Persistent representations are fundamental to modern AI systems, underpinning world models, long-running foundation-model agents, retrieval-augmented systems, and dynamic knowledge structures. These representations are typically treated as static once stored, implicitly assuming that the evidence, context, and task requirements supporting them remain valid. In practice, sensing conditions change, documents are updated, beliefs evolve, and structural relations are revised, causing previously meaningful distinctions to become unsupported or misleading. We argue that persistent AI memory should distinguish an epistemically supportable boundary from a resource-conditioned maintained boundary: the former captures distinctions justified by current evidence and task requirements, whereas the latter captures which of those distinctions can remain explicitly materialized under finite compute, memory, latency, and energy budgets. We introduce boundary-conditioned epistemic memory (BCEM), which jointly maintains an estimated supportable boundary and a resource-feasible maintained boundary. Memory units are organized relative to these boundaries and linked to evidence, provenance, validity, maintainability, lifecycle state, and applicability scope. We formulate memory homeostasis as a closed loop that continuously updates the maintained boundary, reorganizes memory units, and reconfigures their physical realization. This perspective offers a unified framework for designing future AI systems that not only store and retrieve information, but continuously assess which representations remain justified, task-relevant, and resource-feasible.
- 28 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessBoundary-Conditioned Epistemic Memory: A Substrate for Self-Maintaining AI
Persistent representations are fundamental to modern AI systems, underpinning world models, long-running foundation-model agents, retrieval-augmented systems, and dynamic knowledge structures. These representations are typically treated as static once stored, implicitly assuming that the evidence, context, and task requirements supporting them remain valid. In practice, sensing conditions change, documents are updated, beliefs evolve, and structural relations are revised, causing previously meaningful distinctions to become unsupported or misleading. We argue that persistent AI memory should distinguish an epistemically supportable boundary from a resource-conditioned maintained boundary: the former captures distinctions justified by current evidence and task requirements, whereas the latter captures which of those distinctions can remain explicitly materialized under finite compute, memory, latency, and energy budgets. We introduce boundary-conditioned epistemic memory (BCEM), which jointly maintains an estimated supportable boundary and a resource-feasible maintained boundary. Memory units are organized relative to these boundaries and linked to evidence, provenance, validity, maintainability, lifecycle state, and applicability scope. We formulate memory homeostasis as a closed loop that continuously updates the maintained boundary, reorganizes memory units, and reconfigures their physical realization. This perspective offers a unified framework for designing future AI systems that not only store and retrieve information, but continuously assess which representations remain justified, task-relevant, and resource-feasible.
- Open Access
- 28 August 2026
- Published by Elsevier in SSRN Electronic Journal
Epistemic Memory: A Validity Layer for Self-Maintaining Intelligent Systems
Artificial intelligence has progressively expanded its mechanisms for storing, representing, retrieving, and adapting knowledge. However, existing AI memory paradigms primarily focus on preserving information content, while the validity conditions under which knowledge remains applicable are often treated implicitly. This gap gives rise to a phenomenon we term semantic coordinate drift: when an embodied agent moves, changes sensors, or encounters novel environmental configurations, previously valid semantic distinctions may become unsupported while new distinctions emerge, yet the agent's internal representation remains fixed. This paper suggests that the evolution toward self-maintaining intelligent systems motivates an emerging memory paradigm-epistemic memory. Rather than replacing existing memory mechanisms, epistemic memory provides a validity-maintenance layer that governs when stored knowledge remains applicable. Here, self-maintaining intelligence refers to the capability of preserving and reorganizing the validity of knowledge representations under changing observation conditions. Epistemic memory explicitly tracks not only what is known, but also the boundary of knowability under current observation conditions. We formalize the dynamic epistemic quotient, an observation-induced equivalence structure over hypotheses that evolves with an agent's interaction history and sensing capabilities. We derive a pairwise incompatibility lower bound showing that any fixed semantic representation must incur unavoidable error across changing epistemic boundaries. We further identify a failure mode under symmetric crossing quotients, where overlap-based class-level transport becomes independent of pre-transition belief, motivating the preservation of within-class hypothesis provenance. We introduce Observable Belief Memory (OBM) as a constructive epistemic governance architecture, with state Mt = (Qt , bt ,ρt), combining the current epistemic quotient, belief over quotient classes, and within-class provenance through a posterior-consistent update. Controlled experiments demonstrate that explicit epistemic tracking improves robustness under changing observation conditions, particularly when class-level correspondence becomes insufficient. These results suggest that self-maintaining intelligent systems may benefit from reasoning not only about hidden states, but also about the evolving validity of the representations through which those states become knowable.
- 28 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessEpistemic Memory: A Validity Layer for Self-Maintaining Intelligent Systems
Artificial intelligence has progressively expanded its mechanisms for storing, representing, retrieving, and adapting knowledge. However, existing AI memory paradigms primarily focus on preserving information content, while the validity conditions under which knowledge remains applicable are often treated implicitly. This gap gives rise to a phenomenon we term semantic coordinate drift: when an embodied agent moves, changes sensors, or encounters novel environmental configurations, previously valid semantic distinctions may become unsupported while new distinctions emerge, yet the agent's internal representation remains fixed. This paper suggests that the evolution toward self-maintaining intelligent systems motivates an emerging memory paradigm-epistemic memory. Rather than replacing existing memory mechanisms, epistemic memory provides a validity-maintenance layer that governs when stored knowledge remains applicable. Here, self-maintaining intelligence refers to the capability of preserving and reorganizing the validity of knowledge representations under changing observation conditions. Epistemic memory explicitly tracks not only what is known, but also the boundary of knowability under current observation conditions. We formalize the dynamic epistemic quotient, an observation-induced equivalence structure over hypotheses that evolves with an agent's interaction history and sensing capabilities. We derive a pairwise incompatibility lower bound showing that any fixed semantic representation must incur unavoidable error across changing epistemic boundaries. We further identify a failure mode under symmetric crossing quotients, where overlap-based class-level transport becomes independent of pre-transition belief, motivating the preservation of within-class hypothesis provenance. We introduce Observable Belief Memory (OBM) as a constructive epistemic governance architecture, with state Mt = (Qt , bt ,ρt), combining the current epistemic quotient, belief over quotient classes, and within-class provenance through a posterior-consistent update. Controlled experiments demonstrate that explicit epistemic tracking improves robustness under changing observation conditions, particularly when class-level correspondence becomes insufficient. These results suggest that self-maintaining intelligent systems may benefit from reasoning not only about hidden states, but also about the evolving validity of the representations through which those states become knowable.
- 8 August 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 351
Observable quotient world models for knowledge-state abstraction under partial observability
- 8 August 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 351
Observable quotient world models for knowledge-state abstraction under partial observability
- 28 July 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 351
Query-conditioned knowledge alignment for cross-system medical reasoning
- 28 July 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 351
Query-conditioned knowledge alignment for cross-system medical reasoning
- Open Access
- 24 June 2026
- Published by Elsevier in SSRN Electronic Journal
Observable Quotient World Models for Knowledge-State Abstraction under Partial Observability
Partially observable intelligent systems require knowledge-state representations that preserve observable state distinctions while suppressing nuisance variation unsupported by the sensing process. Existing latent world models learn useful predictive states, but their representation targets are often shaped by reconstruction, prediction, task supervision, or architectural bias rather than by observable distinguishability. This paper introduces Observable Quotient Theory, a sensing-induced state-abstraction framework defined through action-conditioned canonicalized observation laws. States indistinguishable under the admissible sensing-action family form observable quotient classes, inducing quotient beliefs, quotient-level dynamics under closure assumptions, recursive filtering, and quotient information gain (QIG) for action selection. We further propose the Quotient World Model (QWM), a neural world model inspired by this theory that learns quotient-relevant representations, beliefs, canonicalized predictions, transition fits, and QIG-based action scores under fixed benchmark and passive-proxy protocols, without identifying learned latents with formal quotient states. A controlled non-radar Gaussian benchmark, COGS, isolates the observable-quotient mechanism and shows action-family refinement, nuisance suppression, QIG-based entropy reduction, and avoidance of unsupported distinctions under insufficient observability. Controlled QuotientWorld experiments further support benchmark-label representation, belief prediction, transition fitting, ablation, and finite-horizon active-sensing diagnostics. Passive CARRADA and V2X-Radar evaluations provide bounded proxy evidence for representation and dynamics behavior on logged real-radar data, with non-uniform gains across metrics and perturbations. Overall, the results support observable-quotient structure as a practical organizing principle for sensing-induced knowledge-state abstraction and quotient-relevant world modeling under controlled, sequential, and passive-proxy protocols.
- 24 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessObservable Quotient World Models for Knowledge-State Abstraction under Partial Observability
Partially observable intelligent systems require knowledge-state representations that preserve observable state distinctions while suppressing nuisance variation unsupported by the sensing process. Existing latent world models learn useful predictive states, but their representation targets are often shaped by reconstruction, prediction, task supervision, or architectural bias rather than by observable distinguishability. This paper introduces Observable Quotient Theory, a sensing-induced state-abstraction framework defined through action-conditioned canonicalized observation laws. States indistinguishable under the admissible sensing-action family form observable quotient classes, inducing quotient beliefs, quotient-level dynamics under closure assumptions, recursive filtering, and quotient information gain (QIG) for action selection. We further propose the Quotient World Model (QWM), a neural world model inspired by this theory that learns quotient-relevant representations, beliefs, canonicalized predictions, transition fits, and QIG-based action scores under fixed benchmark and passive-proxy protocols, without identifying learned latents with formal quotient states. A controlled non-radar Gaussian benchmark, COGS, isolates the observable-quotient mechanism and shows action-family refinement, nuisance suppression, QIG-based entropy reduction, and avoidance of unsupported distinctions under insufficient observability. Controlled QuotientWorld experiments further support benchmark-label representation, belief prediction, transition fitting, ablation, and finite-horizon active-sensing diagnostics. Passive CARRADA and V2X-Radar evaluations provide bounded proxy evidence for representation and dynamics behavior on logged real-radar data, with non-uniform gains across metrics and perturbations. Overall, the results support observable-quotient structure as a practical organizing principle for sensing-induced knowledge-state abstraction and quotient-relevant world modeling under controlled, sequential, and passive-proxy protocols.
- 25 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 39 (4) , 112-119
Enhancing IoMT Security Using Large Multimodal Models
This paper explores the potential of Large Multi-modal Models (LMMs) in enhancing the security of Internet of Medical Things (IoMT) networks. As IoMT systems become more complex and generate vast amounts of sensitive data, traditional security measures are increasingly insufficient to address evolving cyber threats. The ability of LMMs to process and integrate diverse data types, such as sensor data, audio, and video, offers a powerful approach to detecting and mitigating security risks in real time. By focusing on key applications like anomaly detection through traffic prediction and improving Physical Layer Security (PLS), this work demonstrates how LMMs can significantly enhance the resilience of IoMT networks against a wide range of cyber threats. The integration of multimodal data enables more accurate traffic predictions and allows adaptive responses to emerging security challenges. Additionally, LMMs contribute to strengthening PLS by identifying vulnerabilities in communication channels and optimizing resource allocation. Through dynamic data fusion techniques, LMMs ensure that IoMT systems can respond to threats in real-time, while maintaining high security and minimizing impact on system performance.
- 25 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 39 (4) , 112-119
Enhancing IoMT Security Using Large Multimodal Models
This paper explores the potential of Large Multi-modal Models (LMMs) in enhancing the security of Internet of Medical Things (IoMT) networks. As IoMT systems become more complex and generate vast amounts of sensitive data, traditional security measures are increasingly insufficient to address evolving cyber threats. The ability of LMMs to process and integrate diverse data types, such as sensor data, audio, and video, offers a powerful approach to detecting and mitigating security risks in real time. By focusing on key applications like anomaly detection through traffic prediction and improving Physical Layer Security (PLS), this work demonstrates how LMMs can significantly enhance the resilience of IoMT networks against a wide range of cyber threats. The integration of multimodal data enables more accurate traffic predictions and allows adaptive responses to emerging security challenges. Additionally, LMMs contribute to strengthening PLS by identifying vulnerabilities in communication channels and optimizing resource allocation. Through dynamic data fusion techniques, LMMs ensure that IoMT systems can respond to threats in real-time, while maintaining high security and minimizing impact on system performance.