Graph convolutional networks of reconstructed graph structure with constrained Laplacian rank
- PDF / 581,904 Bytes
- 12 Pages / 439.642 x 666.49 pts Page_size
- 11 Downloads / 277 Views
Graph convolutional networks of reconstructed graph structure with constrained Laplacian rank Mengmeng Zhan1 · Jiangzhang Gan2 · Guangquan Lu1 · Yingying Wan1 Received: 22 February 2020 / Revised: 26 August 2020 / Accepted: 24 September 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Convolutional neural networks (CNNs) have achieved unprecedented competitiveness in text and two-dimensional image data processing because of its good accuracy performance and high detection speed. Graph convolutional networks (GCNs), as an extension of classical CNNs in graph data processing, have attracted wide attention. At present, GCNs often use domain knowledge (such as citation recommendation system, biological cell networks) or artificial created fixed graph to achieve various semi-supervised classication tasks. Poor quality graph will lead to suboptimal results of semi-supervised classification tasks. We propose a more general GCN of reconstructed graph structure with constrained Laplacian rank. First, we use hypergraph to establish multivariate relationships between data. On the basis of the hypergraph, In virtue of Laplacian rank constraint to the graph matrix, we learn a new graph structure which has c connected components (where c is the number of classification), and then we construct an ideal graph matrix which is more suitable for the task of semi-supervised classification on GCNs. Finally, the data and the new graph are input GCNs model to get the results of classification. Experiments on 10 different datasets demonstrate that this method is more competitive than the comparison method. Keywords Graph convolutional networks · Adaptive graph · Hypergraph · Semi-supervised classification · Graph structure
1 Introduction At present, deep learning has performed well in solving computer vision [24], image processing [12], speech recognition [1] and other tasks [33]. Among different types of deep learning models, convolutional neural networks (CNNs) have been widely studied and applied because of its high performance in various tasks. LeCun et al. [20] and others Guangquan Lu
[email protected] 1
College of Computer Science and Information Technology, Guangxi Normal University, Guilin, Guangxi, 541004, People’s Republic of China
2
School of Natural and Computational Sciences, Massey University Auckland Campus, Auckland, 0745, New Zealand
Multimedia Tools and Applications
established the basic framework of CNN model inspired by the cognitive mechanism of biological natural vision. Krizhevsky et al. [19] proposed the AlexNet network and won the championship in the 2012 ImageNet, making CNNs become the core algorithm model in image classification. After that, CNNs developed rapidly. Many CNNs models, such as ResNet [13], NasNet [33] and GoogleNet [27], have been proposed by researchers, which show strong recognition ability in target detection, digital classification and other tasks. In particular, although CNNs have good performance and high efficiency, it also has great lim
Data Loading...
