Kui Yu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Kui Yu has a total of 175 co-authors. Here are the top co-authors by number of shared publications:
- Lin Liu
Adelaide University
23 shared publications - Jiuyong Li
Adelaide University
21 shared publications - Fuyuan Cao
School of Computer and Information Technology, Shanxi University, Taiyuan, China
19 shared publications - Xindong Wu
Hefei University of Technology
19 shared publications Xindong Wu
17 shared publicationsWei Ding
University Of Massachusetts Boston, Boston, USA
16 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 2625-2633
SFFedMC: Spurious-Free Federated Multi-View Clustering via Privacy-Preserving Confounder Balancing
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 2625-2633
Open AccessSFFedMC: Spurious-Free Federated Multi-View Clustering via Privacy-Preserving Confounder Balancing
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 38  (11) , 7548-7563
Gravitation and Causality Guided Multi-View Unsupervised Feature Selection
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 38  (11) , 7548-7563
Gravitation and Causality Guided Multi-View Unsupervised Feature Selection
- Open Access
- 19 October 2026
- Published by Elsevier in SSRN Electronic Journal
FLCDNH: Federated Local Causal Discovery for Nonlinear Heterogeneous Data
Federated local causal discovery aims to identify the causal neighborhood of a target variable from distributed data without sharing raw records or recovering a complete causal graph. Existing federated causal discovery methods either focus on global structure learning with exponentially growing search spaces and substantial communication overhead, or restrict to local structures but remain confined to homogeneous linear settings. Real-world federated data, however, often exhibit both data heterogeneity across clients and nonlinear causal mechanisms. To address this gap, we propose FLCDNH, a federated local causal discovery algorithm that is robust to heterogeneity and aware of nonlinearity. FLCDNH incorporates a surrogate domain variable into target parent-child search, spouse candidate expansion, and local skeleton refinement to localize heterogeneity detection. It further introduces a changing-module-aware and abstention-enabled local orientation framework that leverages the domain variable to propagate deterministic orientations, conservative v-structures, and Meek's rules, while evaluating remaining undirected edges using data-type-specific residual-invariance scores. Extensive experiments demonstrate that FLCDNH outperforms state-of-the-art federated global and local causal discovery baselines.
- 19 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessFLCDNH: Federated Local Causal Discovery for Nonlinear Heterogeneous Data
Federated local causal discovery aims to identify the causal neighborhood of a target variable from distributed data without sharing raw records or recovering a complete causal graph. Existing federated causal discovery methods either focus on global structure learning with exponentially growing search spaces and substantial communication overhead, or restrict to local structures but remain confined to homogeneous linear settings. Real-world federated data, however, often exhibit both data heterogeneity across clients and nonlinear causal mechanisms. To address this gap, we propose FLCDNH, a federated local causal discovery algorithm that is robust to heterogeneity and aware of nonlinearity. FLCDNH incorporates a surrogate domain variable into target parent-child search, spouse candidate expansion, and local skeleton refinement to localize heterogeneity detection. It further introduces a changing-module-aware and abstention-enabled local orientation framework that leverages the domain variable to propagate deterministic orientations, conservative v-structures, and Meek's rules, while evaluating remaining undirected edges using data-type-specific residual-invariance scores. Extensive experiments demonstrate that FLCDNH outperforms state-of-the-art federated global and local causal discovery baselines.
- 23 September 2026
- Published by Elsevier in Neural Networks
- Vol. 206  (Pt A) , 109673
Context-aware causal reasoning for explanatory visual question answering
- 23 September 2026
- Published by Elsevier in Neural Networks
- Vol. 206  (Pt A) , 109673
Context-aware causal reasoning for explanatory visual question answering
- Open Access
- 19 September 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
Addressing Span Imbalance and Semantic Complexity in Nested Medical Named Entity Recognition
As a fundamental task in biomedical natural language processing, Medical Named Entity Recognition (MNER) aims to identify and classify medical entities from unstructured medical texts. A major challenge in this task is the prevalence of nested entities, which arise from the syntactic complexity and domain-specific characteristics of medical language. Recently, span-based models have been proposed to handle nested entities by reformulating the task as a multi-label classification problem over all possible spans. However, when directly applied to medical texts, these models encounter two problems. First, existing span-based methods typically rely on general-domain pre-trained language models to learn textual semantics, which hinders their ability to capture contextual dependencies in complex medical texts, thereby limiting their generalization across diverse medical data. Second, these methods treat all candidate spans equally during training, which results in an underestimation of the gradient contributions from entity spans due to their relatively small proportion, thereby degrading recognition performance. To address these issues, we propose AGPNer, a novel method for recognizing nested medical named entities. The proposed AGPNer integrates two key components: (1) a heterogeneous dependency fusion encoder, which reconstructs masked entities to enhance token representations and fine-tunes a hybrid dependency modeling block to learn domain-specific patterns in medical texts; (2) an imbalance-adaptive span decoder, which decouples entity and non-entity spans and adaptively assigns them different exponential decay factors to regulate their contributions during training. Experimental results on four public benchmarks demonstrate that AGPNer achieves absolute improvements in F1-score of 0.90, 1.02, 0.47 and 0.75 percentage points on CMeEE-V1, CMeEE-V2, GENIA, and CLUENER, respectively, showing consistent and competitive performance among the compared baselines.
- 19 September 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
Open AccessAddressing Span Imbalance and Semantic Complexity in Nested Medical Named Entity Recognition
As a fundamental task in biomedical natural language processing, Medical Named Entity Recognition (MNER) aims to identify and classify medical entities from unstructured medical texts. A major challenge in this task is the prevalence of nested entities, which arise from the syntactic complexity and domain-specific characteristics of medical language. Recently, span-based models have been proposed to handle nested entities by reformulating the task as a multi-label classification problem over all possible spans. However, when directly applied to medical texts, these models encounter two problems. First, existing span-based methods typically rely on general-domain pre-trained language models to learn textual semantics, which hinders their ability to capture contextual dependencies in complex medical texts, thereby limiting their generalization across diverse medical data. Second, these methods treat all candidate spans equally during training, which results in an underestimation of the gradient contributions from entity spans due to their relatively small proportion, thereby degrading recognition performance. To address these issues, we propose AGPNer, a novel method for recognizing nested medical named entities. The proposed AGPNer integrates two key components: (1) a heterogeneous dependency fusion encoder, which reconstructs masked entities to enhance token representations and fine-tunes a hybrid dependency modeling block to learn domain-specific patterns in medical texts; (2) an imbalance-adaptive span decoder, which decouples entity and non-entity spans and adaptively assigns them different exponential decay factors to regulate their contributions during training. Experimental results on four public benchmarks demonstrate that AGPNer achieves absolute improvements in F1-score of 0.90, 1.02, 0.47 and 0.75 percentage points on CMeEE-V1, CMeEE-V2, GENIA, and CLUENER, respectively, showing consistent and competitive performance among the compared baselines.
- Open Access
- 23 May 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
- Vol. 44  (5) , 1-23
MGCD: Multiple-Granularity Cognitive Diagnosis in Intelligent Education Systems
Cognitive diagnosis (CD) is an important task in the field of intelligent education, aiming to discover the proficiency of students on knowledge concepts with response logs. In applications, different users of the tutoring system demand for a diagnosis of knowledge concepts at different granularities. However, recent methods assume that the concepts are of the same granularity and use explicit correlations between same-granularity concepts to improve the diagnosis performance. If required for diagnosing multi-granularity concepts, these methods will face diminished performance or partial invalidation. To this end, we make the first attempt for multiple-granularity cognitive diagnosis, i.e., diagnosis on coarse- and fine-grained concepts simultaneously. Specifically, in a skillful way, the same-granularity correlations are captured and embedded into concept representations in view of concept semantics and cross-granularity correlations to model the proficiency influence between concepts implicitly. Then, the specific loss for single-granularity diagnosis and the general loss for the consistency of multi-granularity are designed to train the model jointly, achieving multiple-granularity diagnosis. Extensive experiments demonstrate that our method can achieve state-of-the-art accuracy on both coarse- and fine-grained concepts.
- 23 May 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
- Vol. 44  (5) , 1-23
Open AccessMGCD: Multiple-Granularity Cognitive Diagnosis in Intelligent Education Systems
Cognitive diagnosis (CD) is an important task in the field of intelligent education, aiming to discover the proficiency of students on knowledge concepts with response logs. In applications, different users of the tutoring system demand for a diagnosis of knowledge concepts at different granularities. However, recent methods assume that the concepts are of the same granularity and use explicit correlations between same-granularity concepts to improve the diagnosis performance. If required for diagnosing multi-granularity concepts, these methods will face diminished performance or partial invalidation. To this end, we make the first attempt for multiple-granularity cognitive diagnosis, i.e., diagnosis on coarse- and fine-grained concepts simultaneously. Specifically, in a skillful way, the same-granularity correlations are captured and embedded into concept representations in view of concept semantics and cross-granularity correlations to model the proficiency influence between concepts implicitly. Then, the specific loss for single-granularity diagnosis and the general loss for the consistency of multi-granularity are designed to train the model jointly, achieving multiple-granularity diagnosis. Extensive experiments demonstrate that our method can achieve state-of-the-art accuracy on both coarse- and fine-grained concepts.
- 20 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. PP  (5) , 1-12
Invariant Subgraph Learning via Effective Environment Subgraph Augmentation for Generalizable Graph Neural Network
With the increasing prevalence of graph data in various practical applications, Graph Neural Networks (GNNs) have established themselves as essential tools for effective graph data processing. However, existing GNNs always perform well on in-distribution data, but exhibit significant performance degradation under distribution shifts. To improve GNN generalization, graph invariant learning aims to identify accurate invariant subgraphs but is constrained by limited diversity in environment subgraphs. Conversely, graph data augmentation focuses on enriching this diversity through environment subgraph augmentation, but its efficacy heavily depends on having accurate invariant subgraphs first. This creates a core paradox: acquiring accurate invariant subgraphs requires diverse data, whereas effective augmentation presupposes accurate invariant subgraphs. To address this issue, we propose IGESA, a method for learning accurate Invariant subGraph via effective Environment Subgraph Augmentation. IGESA introduces two key strategies: (1) For accurate invariant subgraph identification, we propose a precise invariant subgraph extraction strategy to refine the subgraph learning process. (2) For sufficiently diverse augmentations, we propose a cross-graph environment fusion strategy that combines sampled components from pair-wise distinct environment subgraphs to construct new subgraphs. These two strategies are collaboratively optimized to boost GNNs' generalizability, and extensive experiments on benchmark datasets demonstrate their superiority over state-of-the-art methods.
- 20 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. PP  (5) , 1-12
Invariant Subgraph Learning via Effective Environment Subgraph Augmentation for Generalizable Graph Neural Network
With the increasing prevalence of graph data in various practical applications, Graph Neural Networks (GNNs) have established themselves as essential tools for effective graph data processing. However, existing GNNs always perform well on in-distribution data, but exhibit significant performance degradation under distribution shifts. To improve GNN generalization, graph invariant learning aims to identify accurate invariant subgraphs but is constrained by limited diversity in environment subgraphs. Conversely, graph data augmentation focuses on enriching this diversity through environment subgraph augmentation, but its efficacy heavily depends on having accurate invariant subgraphs first. This creates a core paradox: acquiring accurate invariant subgraphs requires diverse data, whereas effective augmentation presupposes accurate invariant subgraphs. To address this issue, we propose IGESA, a method for learning accurate Invariant subGraph via effective Environment Subgraph Augmentation. IGESA introduces two key strategies: (1) For accurate invariant subgraph identification, we propose a precise invariant subgraph extraction strategy to refine the subgraph learning process. (2) For sufficiently diverse augmentations, we propose a cross-graph environment fusion strategy that combines sampled components from pair-wise distinct environment subgraphs to construct new subgraphs. These two strategies are collaboratively optimized to boost GNNs' generalizability, and extensive experiments on benchmark datasets demonstrate their superiority over state-of-the-art methods.
- Open Access
- 27 April 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17  (3) , 1-19
PMARL: Multi-Agent Reinforcement Learning in Large-Scale Systems
Large-scale multi-agent systems face two core challenges: inefficient policy learning and the explosion of state dimensions. Existing methods often rely on manually designed task sequences to guide agents’ learning in stages, but these designs lack adaptability to agents’ learning abilities, making it difficult to ensure the rationality of task difficulty. Moreover, the representation capability of current network structures is limited, making it challenging to efficiently handle high-dimensional state information and complex interaction relationships. To address these issues, we propose a Progressive Multi-Agent Reinforcement Learning (PMARL) framework. PMARL introduces a task adapter that adaptively selects task difficulty based on agents’ learning abilities, eliminating reliance on manual experience. Additionally, a Dynamic Dimension Adaptive Network (DDAN) is designed, incorporating hypernetwork and self-attention mechanisms to achieve adaptive feature extraction of high-dimensional states and efficient representation of agent interaction relationships. Experimental results demonstrate that PMARL exhibits higher efficiency and better adaptability compared to existing methods when addressing large-scale multi-agent tasks.
- 27 April 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17  (3) , 1-19
Open AccessPMARL: Multi-Agent Reinforcement Learning in Large-Scale Systems
Large-scale multi-agent systems face two core challenges: inefficient policy learning and the explosion of state dimensions. Existing methods often rely on manually designed task sequences to guide agents’ learning in stages, but these designs lack adaptability to agents’ learning abilities, making it difficult to ensure the rationality of task difficulty. Moreover, the representation capability of current network structures is limited, making it challenging to efficiently handle high-dimensional state information and complex interaction relationships. To address these issues, we propose a Progressive Multi-Agent Reinforcement Learning (PMARL) framework. PMARL introduces a task adapter that adaptively selects task difficulty based on agents’ learning abilities, eliminating reliance on manual experience. Additionally, a Dynamic Dimension Adaptive Network (DDAN) is designed, incorporating hypernetwork and self-attention mechanisms to achieve adaptive feature extraction of high-dimensional states and efficient representation of agent interaction relationships. Experimental results demonstrate that PMARL exhibits higher efficiency and better adaptability compared to existing methods when addressing large-scale multi-agent tasks.
- Open Access
- 27 April 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17  (3) , 1-32
Mutual Information-Guided Style Augmentation for Single Domain Generalization
Single domain generalization aims to develop a robust model trained on a source domain to generalize well on unseen target domains. Recent progress in single domain generalization has focused on expanding the scope of training data through style (e.g., backgrounds) augmentation. However, existing methods are difficult to generate data with large style shifts due to the lack of precise correlation measures between the generated and original data, and they struggle to effectively capture the consistency between the generated and original data when learning feature representations. In this article, we propose a novel Mutual Information-guided Style Augmentation (MISA) based single domain generalization method. Specifically, MISA incorporates a style diversity module, which uses the matrix-based Rényi’s \(\alpha\) -order entropy functionals to compute an approximate mutual information value between the augmented and original data, minimizing it to guide style generator learning. Moreover, MISA combines the merits of the random convolution and affine transformation to further improve the texture diversity of the augmented data. Additionally, MISA introduces a representation learning module, which minimizes the approximate mutual information value between the prediction logits of the original sample and its corresponding residual component to capture the consistency between the generated and original data for feature representation optimization. Using five real-world datasets, the extensive experiments have demonstrated the effectiveness of MISA, in comparison with state-of-the-art methods.- 27 April 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17  (3) , 1-32
Open AccessMutual Information-Guided Style Augmentation for Single Domain Generalization
Single domain generalization aims to develop a robust model trained on a source domain to generalize well on unseen target domains. Recent progress in single domain generalization has focused on expanding the scope of training data through style (e.g., backgrounds) augmentation. However, existing methods are difficult to generate data with large style shifts due to the lack of precise correlation measures between the generated and original data, and they struggle to effectively capture the consistency between the generated and original data when learning feature representations. In this article, we propose a novel Mutual Information-guided Style Augmentation (MISA) based single domain generalization method. Specifically, MISA incorporates a style diversity module, which uses the matrix-based Rényi’s \(\alpha\) -order entropy functionals to compute an approximate mutual information value between the augmented and original data, minimizing it to guide style generator learning. Moreover, MISA combines the merits of the random convolution and affine transformation to further improve the texture diversity of the augmented data. Additionally, MISA introduces a representation learning module, which minimizes the approximate mutual information value between the prediction logits of the original sample and its corresponding residual component to capture the consistency between the generated and original data for feature representation optimization. Using five real-world datasets, the extensive experiments have demonstrated the effectiveness of MISA, in comparison with state-of-the-art methods.