期刊
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
卷 32, 期 10, 页码 4755-4760出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TNNLS.2020.3025110
关键词
Laplace equations; Convolutional codes; Protocols; Neural networks; Learning systems; Adaptive systems; Adaptation models; Convolutional network; graph data; graph neural network (GNN); node classification
This brief introduces an adaptive propagation GCN model that achieves superior or similar results to the best proposed models by adjusting the number of communication steps independently at every node. Additionally, a regularization term is investigated to enforce an explicit tradeoff between communication and accuracy. The code for the AP-GCN experiments has been released as an open-source library.
Graph convolutional networks (GCNs) are a family of neural network models that perform inference on graph data by interleaving vertexwise operations and message-passing exchanges across nodes. Concerning the latter, two key questions arise: 1) how to design a differentiable exchange protocol (e.g., a one-hop Laplacian smoothing in the original GCN) and 2) how to characterize the tradeoff in complexity with respect to the local updates. In this brief, we show that the state-of-the-art results can be achieved by adapting the number of communication steps independently at every node. In particular, we endow each node with a halting unit (inspired by Graves' adaptive computation time [1]) that after every exchange decides whether to continue communicating or not. We show that the proposed adaptive propagation GCN (AP-GCN) achieves superior or similar results to the best proposed models so far on a number of benchmarks while requiring a small overhead in terms of additional parameters. We also investigate a regularization term to enforce an explicit tradeoff between communication and accuracy. The code for the AP-GCN experiments is released as an open-source library.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据