Author name cluster

Weigang Li

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

3 papers

1 author row

EAAI Journal 2024 Journal Article

An infill sampling criterion based on improvement of probability and mapping crowding distance for expensive multi/many-objective optimization

Yang Li
Weigang Li
Songtao Li
Yuntao Zhao

Over the preceding decade, surrogate-assisted multi-objective optimization has attracted increasing attention, especially for optimization problems involving time-consuming simulations or expensive physical experiments. In this work, an infill sampling criterion based on the improvement of probability and mapping crowding distance (PIMD) is proposed for expensive multi/many-objective optimization. First, the improvement matrix of probability is derived from probability of improvement (PI), from which a scalar indicator is aggregated to comprehensively evaluate the convergence and uncertainty of trial solutions. Then, leveraging the estimated geometric shape of the Pareto front, the relevant points are mapped onto the inferred surface to effectively assess diversity. To maintain the balance between convergence, diversity, and uncertainty of trial solutions, the infill sampling task is transformed into an optimization problem considering the above two indicators, and the corresponding Pareto solutions are selected for re-evaluation using expensive function evaluations. Numerical experiments have been conducted on 90 standard instances and 2 real-world problems. Comparative analysis with eight state-of-the-art algorithms indicates that the proposed PIMD achieves competitive performance on the majority of the involved problems.

Details DOI

EAAI Journal 2024 Journal Article

Cross-domain few-shot fault diagnosis based on meta-learning and domain adversarial graph convolutional network

Junwei Hu
Weigang Li
Yong Zhang
Zhiqiang Tian

Problems such as small samples and variable working conditions arise in complex practical mechanical fault diagnosis scenarios. Although the domain-adaptive method has an excellent effect in the task of domain deviation diagnosis, it still needs a large number of source domain marker samples. Therefore, a new meta-learning domain-adversarial graph convolutional network (MDGCN) is proposed to solve the above problem of practical fault diagnosis for cross-domain few-shot (CDFS). Firstly, the graph convolutional network is used to learn the relationship between the structural features of the data from the input signal to construct the instance graph. Then, domain-adversarial training is used to minimize the domain discrimination error of source and target data for domain adaptation. Simultaneously, a scaled distance metric function is used to compute the similarity between the faulty sample and the sample prototype for fault diagnosis. Finally, the MDGCN model is evaluated using CDFS cases from three mechanical vibration datasets. Experimental results show that the proposed method not only achieves the best performance in comparison, but also extracts the features of CDFS faults for domain-adaptive diagnosis.

Details DOI

EAAI Journal 2023 Journal Article

Semi-supervised non-negative matrix tri-factorization with adaptive neighbors and block-diagonal learning

Songtao Li
Weigang Li
Hao Lu
Yang Li

Graph-regularized non-negative matrix factorization (GNMF) is proved to be effective for the clustering of nonlinear separable data. Existing GNMF variants commonly improve model performance by adding different additional constraints or refining the model factorization form, which can lead to problems such as increased algorithm complexity or insufficient performance release. In this paper, we propose semi-supervised non-negative matrix tri-factorization with adaptive neighbors and block-diagonal (ABNMTF). Different from existing methods, in ABNMTF the similarity graph matrix is learned from the original data by adaptive neighbors k-nearest model, and a block diagonal matrix is constructed based on a few labeled data to update the similarity matrix. Our approach reconstructs the block diagonal structure into the adaptive similarity matrix, which enables simultaneous learning of the similarity matrix and label binding during factorization, engendering a distinguishable subspace representation matrix and therefore improving the clustering performance without significantly increasing the complexity of the algorithm. We also represent an optimization method to solve the ABNMTF and provide analyses of convergence and computational complexity. Extensive experiments on 8 real image datasets show that the proposed algorithm reports superior performance against several state-of-the-art approaches. Code has been made available at: https: //github. com/LstinWh/ABNMTF.

Details DOI