HiLNO: A Hierarchical Latent Neural Operator with Multi-Scale Supervision for PDEs on General Geometries
为解决PDE解的多尺度结构信息丢失问题，提出HiLNO方法，通过构建细到粗再到细的潜在空间，并引入多尺度监督和各向异性高斯注意力机制。
0 citationsRead paper
Selected work
Representative Papers
BRkNN-light: Batch Processing of Reverse k-Nearest Neighbor Queries for Moving Objects on Road Networks
To address the inefficiency caused by redundant computations in concurrent reverse k-nearest neighbor (RkNN) queries over road networks, this paper proposes the first batch-processing framework for RkNN queries tailored to road networks. Methodologically, it introduces a geometry-constrained pruning and verification mechanism to significantly reduce the search space; employs a dynamic distance cache to enable cross-query reuse of distance computations; and integrates a batch-wise collaborative optimization strategy to improve overall throughput. Extensive experiments on multiple real-world road network datasets demonstrate that the proposed approach reduces total computational cost by up to several-fold compared to state-of-the-art single-query methods, while substantially decreasing query latency. This work achieves, for the first time, efficient and scalable batch RkNN processing in road-network environments.
0 citationsRead paper
More Consistent Accuracy PINN via Alternating Easy-Hard Training
To address the training instability, large accuracy fluctuations, and poor generalization across diverse problems inherent in Physics-Informed Neural Networks (PINNs) for solving partial differential equations (PDEs), this work proposes an Alternating Easy–Hard Sample Prioritization Training strategy. The method jointly couples hard-first prioritization—driven by residuals and gradients—with easy-first prioritization—guided by low-frequency components and initial/boundary conditions. We introduce, for the first time, an alternating optimization framework integrating residual-weighted sampling, frequency-domain analysis, and dynamic loss reweighting, augmented with automatic differentiation and adaptive mesh redistribution. This approach overcomes the limitations of single-priority strategies, achieving relative L² errors of 1e−5–1e−6 on PDEs with steep gradients, strong nonlinearity, and high dimensionality. It consistently outperforms state-of-the-art baselines by 3–10× in accuracy, while significantly enhancing robustness and generalization consistency.
0 citationsRead paper
Hybrid Matrix Factorization Based Graph Contrastive Learning for Recommendation System
To address data sparsity and insufficient interaction modeling in graph-based contrastive learning for recommendation, this paper proposes a hybrid matrix factorization framework integrating low-rank matrix factorization (MF) and singular value decomposition (SVD) to generate augmented views enriched with global collaborative signals—replacing conventional data augmentation strategies reliant on graph structural perturbation or clustering. MF captures high-order collaborative patterns, while SVD extracts latent semantic structures; their complementary modeling of user–item interactions enhances the discriminative power of self-supervised graph neural network training. Extensive experiments on multiple public benchmarks demonstrate that the proposed method significantly outperforms state-of-the-art baselines, particularly under sparse-data conditions, where it achieves substantial gains in recommendation accuracy. These results validate its effectiveness in mitigating data sparsity and improving robustness in collaborative filtering.
0 citationsRead paper
Contextual Preference Collaborative Measure Framework Based on Belief System
To address the challenges of excessive human intervention, low accuracy, and poor efficiency in preference measurement, this paper proposes a context-aware collaborative preference measurement framework based on a dynamic belief system. Methodologically, it introduces— for the first time—the “shared belief” update mechanism coupled with the Preference Refinement Algorithm (PRA) to enable low-information-loss extraction of common preferences. It further constructs an unsupervised, dual-dimensional rule discrimination model grounded in belief strength and bias magnitude, supporting automatic generalization and personalization of preference classification. Additionally, it designs extensible interestingness measures, including IMCos (weighted cosine similarity) and IMCov (correlation coefficient). Experimental results demonstrate that the proposed approach significantly outperforms two state-of-the-art baselines across accuracy, runtime efficiency, and Top-K rule quality.
0 citationsRead paper
Recent publications
Latest Papers
HiLNO: A Hierarchical Latent Neural Operator with Multi-Scale Supervision for PDEs on General Geometries
为解决PDE解的多尺度结构信息丢失问题，提出HiLNO方法，通过构建细到粗再到细的潜在空间，并引入多尺度监督和各向异性高斯注意力机制。
0 citationsRead paper
BRkNN-light: Batch Processing of Reverse k-Nearest Neighbor Queries for Moving Objects on Road Networks
To address the inefficiency caused by redundant computations in concurrent reverse k-nearest neighbor (RkNN) queries over road networks, this paper proposes the first batch-processing framework for RkNN queries tailored to road networks. Methodologically, it introduces a geometry-constrained pruning and verification mechanism to significantly reduce the search space; employs a dynamic distance cache to enable cross-query reuse of distance computations; and integrates a batch-wise collaborative optimization strategy to improve overall throughput. Extensive experiments on multiple real-world road network datasets demonstrate that the proposed approach reduces total computational cost by up to several-fold compared to state-of-the-art single-query methods, while substantially decreasing query latency. This work achieves, for the first time, efficient and scalable batch RkNN processing in road-network environments.
0 citationsRead paper
More Consistent Accuracy PINN via Alternating Easy-Hard Training
To address the training instability, large accuracy fluctuations, and poor generalization across diverse problems inherent in Physics-Informed Neural Networks (PINNs) for solving partial differential equations (PDEs), this work proposes an Alternating Easy–Hard Sample Prioritization Training strategy. The method jointly couples hard-first prioritization—driven by residuals and gradients—with easy-first prioritization—guided by low-frequency components and initial/boundary conditions. We introduce, for the first time, an alternating optimization framework integrating residual-weighted sampling, frequency-domain analysis, and dynamic loss reweighting, augmented with automatic differentiation and adaptive mesh redistribution. This approach overcomes the limitations of single-priority strategies, achieving relative L² errors of 1e−5–1e−6 on PDEs with steep gradients, strong nonlinearity, and high dimensionality. It consistently outperforms state-of-the-art baselines by 3–10× in accuracy, while significantly enhancing robustness and generalization consistency.
0 citationsRead paper
Hybrid Matrix Factorization Based Graph Contrastive Learning for Recommendation System
To address data sparsity and insufficient interaction modeling in graph-based contrastive learning for recommendation, this paper proposes a hybrid matrix factorization framework integrating low-rank matrix factorization (MF) and singular value decomposition (SVD) to generate augmented views enriched with global collaborative signals—replacing conventional data augmentation strategies reliant on graph structural perturbation or clustering. MF captures high-order collaborative patterns, while SVD extracts latent semantic structures; their complementary modeling of user–item interactions enhances the discriminative power of self-supervised graph neural network training. Extensive experiments on multiple public benchmarks demonstrate that the proposed method significantly outperforms state-of-the-art baselines, particularly under sparse-data conditions, where it achieves substantial gains in recommendation accuracy. These results validate its effectiveness in mitigating data sparsity and improving robustness in collaborative filtering.
0 citationsRead paper
Contextual Preference Collaborative Measure Framework Based on Belief System
To address the challenges of excessive human intervention, low accuracy, and poor efficiency in preference measurement, this paper proposes a context-aware collaborative preference measurement framework based on a dynamic belief system. Methodologically, it introduces— for the first time—the “shared belief” update mechanism coupled with the Preference Refinement Algorithm (PRA) to enable low-information-loss extraction of common preferences. It further constructs an unsupervised, dual-dimensional rule discrimination model grounded in belief strength and bias magnitude, supporting automatic generalization and personalization of preference classification. Additionally, it designs extensible interestingness measures, including IMCos (weighted cosine similarity) and IMCov (correlation coefficient). Experimental results demonstrate that the proposed approach significantly outperforms two state-of-the-art baselines across accuracy, runtime efficiency, and Top-K rule quality.
0 citationsRead paper