GRACE: Adaptive Concept Erasure with Geometry-Guided Retention in Diffusion Models
为解决文本到图像扩散模型中的敏感概念删除问题，提出GRACE框架，通过几何引导的保留和自适应干预方法，在保证生成质量的同时有效移除不合规内容。
0 citationsRead paper
Selected work
Representative Papers
Dynamic Trust-Aware Sparse Communication Topology for LLM-Based Multi-Agent Consensus
This work addresses the scalability limitations of existing large language model–based multi-agent systems, where fully connected communication incurs near-quadratic growth in communication overhead with agent count, while fixed sparse topologies lack adaptability to dynamic task demands. The authors propose DySCo, a novel dynamic trust-aware sparse communication mechanism that, for each reasoning round, evaluates the utility of potential communication edges based on agent reliability, answer divergence, and task relevance. Under a communication budget constraint, DySCo selectively activates high-value edges for on-demand interaction and aggregates responses using dynamically adjusted trust weights, terminating discussions early upon achieving consensus stability. Experiments demonstrate that DySCo substantially reduces communication overhead and latency across mathematical reasoning, logical reasoning, and factual question-answering tasks, while maintaining or even improving accuracy and consensus stability.
0 citationsRead paper
Combined-distance-based score function of cognitive fuzzy sets and its application in lung cancer pain evaluation
Existing distance measures for Cognitive Fuzzy Sets (CFS) are insufficient, particularly the Minkowski distance, which neglects hesitation degree and thereby introduces decision bias. Method: This paper pioneers the integration of Hausdorff distance principles into CFS, proposing an improved Minkowski distance (CF-IM) and a novel Hausdorff-based distance (CF-H). A robust composite distance (CF-C) is then constructed via linear weighting to jointly ensure noise resistance and information completeness; a new scoring function is derived therefrom. Contribution/Results: Evaluated on lung cancer pain assessment, the proposed method demonstrates superior sensitivity, stability, and discriminative power. Comparative experiments confirm its reliability and superiority over conventional CFS distances, significantly enhancing modeling accuracy for complex cognitive uncertainty and strengthening decision-support capability.
0 citationsRead paper
Attributed Graph Clustering with Multi-Scale Weight-Based Pairwise Coarsening and Contrastive Learning
Existing attributed graph clustering methods suffer from three key challenges: inadequate modeling of long-range dependencies, feature collapse, and structural information loss during graph coarsening. To address these, this paper proposes a multi-scale weighted dual coarsening framework integrated with one-to-many contrastive learning. First, a global similarity-guided edge merging strategy is designed to preserve both long-range dependencies and fine-grained structural details during coarsening. Second, a cluster-center-aware one-to-many contrastive learning objective is introduced to mitigate feature masking caused by high-degree nodes and enhance representation diversity. Third, a multi-scale weight-aware coarsening mechanism is jointly optimized with a graph reconstruction loss and KL-divergence regularization to ensure structural consistency across scales. Extensive experiments on ACM, Citeseer, Cora, and DBLP demonstrate significant and robust performance gains, with the normalized mutual information (NMI) improving by up to 15.24%.
0 citationsRead paper
GCL-GCN: Graphormer and Contrastive Learning Enhanced Attributed Graph Clustering Network
Addressing challenges in attributed graph clustering—namely, insufficient local dependency modeling and difficulty capturing global structure due to graph sparsity and node attribute heterogeneity—this paper proposes a deep graph clustering framework integrating global and local representation learning. Its core contributions are: (1) a centrality-enhanced spatial attention Graphormer module that jointly encodes node centrality and spatial relationships, explicitly modeling both long-range dependencies and local neighborhood structures; and (2) a contrastive learning-based two-stage pretraining strategy to enhance node representation discriminability and clustering robustness. The framework jointly optimizes representation learning and clustering objectives in an end-to-end manner. Extensive experiments on six benchmark datasets demonstrate significant improvements over 14 state-of-the-art methods: on Cora, it achieves absolute gains of 4.94% in ACC, 13.01% in NMI, and 10.97% in ARI, validating its effectiveness and stability.
0 citationsRead paper
Recent publications
Latest Papers
GRACE: Adaptive Concept Erasure with Geometry-Guided Retention in Diffusion Models
为解决文本到图像扩散模型中的敏感概念删除问题，提出GRACE框架，通过几何引导的保留和自适应干预方法，在保证生成质量的同时有效移除不合规内容。
0 citationsRead paper
Dynamic Trust-Aware Sparse Communication Topology for LLM-Based Multi-Agent Consensus
This work addresses the scalability limitations of existing large language model–based multi-agent systems, where fully connected communication incurs near-quadratic growth in communication overhead with agent count, while fixed sparse topologies lack adaptability to dynamic task demands. The authors propose DySCo, a novel dynamic trust-aware sparse communication mechanism that, for each reasoning round, evaluates the utility of potential communication edges based on agent reliability, answer divergence, and task relevance. Under a communication budget constraint, DySCo selectively activates high-value edges for on-demand interaction and aggregates responses using dynamically adjusted trust weights, terminating discussions early upon achieving consensus stability. Experiments demonstrate that DySCo substantially reduces communication overhead and latency across mathematical reasoning, logical reasoning, and factual question-answering tasks, while maintaining or even improving accuracy and consensus stability.
0 citationsRead paper
Combined-distance-based score function of cognitive fuzzy sets and its application in lung cancer pain evaluation
Existing distance measures for Cognitive Fuzzy Sets (CFS) are insufficient, particularly the Minkowski distance, which neglects hesitation degree and thereby introduces decision bias. Method: This paper pioneers the integration of Hausdorff distance principles into CFS, proposing an improved Minkowski distance (CF-IM) and a novel Hausdorff-based distance (CF-H). A robust composite distance (CF-C) is then constructed via linear weighting to jointly ensure noise resistance and information completeness; a new scoring function is derived therefrom. Contribution/Results: Evaluated on lung cancer pain assessment, the proposed method demonstrates superior sensitivity, stability, and discriminative power. Comparative experiments confirm its reliability and superiority over conventional CFS distances, significantly enhancing modeling accuracy for complex cognitive uncertainty and strengthening decision-support capability.
0 citationsRead paper
Attributed Graph Clustering with Multi-Scale Weight-Based Pairwise Coarsening and Contrastive Learning
Existing attributed graph clustering methods suffer from three key challenges: inadequate modeling of long-range dependencies, feature collapse, and structural information loss during graph coarsening. To address these, this paper proposes a multi-scale weighted dual coarsening framework integrated with one-to-many contrastive learning. First, a global similarity-guided edge merging strategy is designed to preserve both long-range dependencies and fine-grained structural details during coarsening. Second, a cluster-center-aware one-to-many contrastive learning objective is introduced to mitigate feature masking caused by high-degree nodes and enhance representation diversity. Third, a multi-scale weight-aware coarsening mechanism is jointly optimized with a graph reconstruction loss and KL-divergence regularization to ensure structural consistency across scales. Extensive experiments on ACM, Citeseer, Cora, and DBLP demonstrate significant and robust performance gains, with the normalized mutual information (NMI) improving by up to 15.24%.
0 citationsRead paper
GCL-GCN: Graphormer and Contrastive Learning Enhanced Attributed Graph Clustering Network
Addressing challenges in attributed graph clustering—namely, insufficient local dependency modeling and difficulty capturing global structure due to graph sparsity and node attribute heterogeneity—this paper proposes a deep graph clustering framework integrating global and local representation learning. Its core contributions are: (1) a centrality-enhanced spatial attention Graphormer module that jointly encodes node centrality and spatial relationships, explicitly modeling both long-range dependencies and local neighborhood structures; and (2) a contrastive learning-based two-stage pretraining strategy to enhance node representation discriminability and clustering robustness. The framework jointly optimizes representation learning and clustering objectives in an end-to-end manner. Extensive experiments on six benchmark datasets demonstrate significant improvements over 14 state-of-the-art methods: on Cora, it achieves absolute gains of 4.94% in ACC, 13.01% in NMI, and 10.97% in ARI, validating its effectiveness and stability.
0 citationsRead paper