Deep Multi-attributed Graph Translation with Node-Edge Co-Evolution

Abstract

Generalized from image and language translation, graph translation aims to generate a graph in the target domain by conditioning an input graph in the source domain. This promising topic has attracted fast-increasing attentions recently. Existing works are limited to either merely predicting the node attributes of graphs with fixed topology or predicting only the graph topology without considering node attributes, but cannot simultaneously predict both of them, due to substantial challenges: 1) difficulty in characterizing the interactive, iterative, and asynchronous translation process of both nodes and edges and 2) difficulty in discovering and maintaining the inherent consistency between the node and edge in predicted graphs. These challenges prevent a generic, end-to-end framework for joint node and edge attributes prediction, which is a need for realworld applications such as malware confinement in IoT networks and structural-to-functional network translation. These realworld applications highly depend on hand-crafting and ad-hoc heuristic models, but cannot sufficiently utilize massive historical data. In this paper, we termed this generic problem "multiattributed graph translation" and developed a novel framework integrating both node and edge translations seamlessly. The novel edge translation path is generic, which is proven to be a generalization of the existing topology translation models. Then, a spectral graph regularization based on our non-parametric graph Laplacian is proposed in order to learn and maintain the consistency of the predicted nodes and edges. Finally, extensive experiments on both synthetic and real-world application data demonstrated the effectiveness of the proposed method.

Graph Neural Networks for Natural Language Processing: A Survey

“…Encoder-decoder architecture is one of the most widely used machine learning framework in the NLP field, such as the Sequence-to-Sequence (Seq2Seq) models (Sutskever et al, 2014;Cho et al, 2014). Given the great power of GNNs for modeling graph-structured data, very recently, many research efforts have been made to develop GNN-based encoder-docoder frameworks including Graph-to-Tree and Graph-to-Graph (Guo et al, 2019a;Shi et al, 2020) models. In this section, we will first introduce the typical Seq2Seq models, and then discuss various graph-based encoder-decoder models for various NLP tasks.…”

“…The goal of graph transformation is to transform an input graph in the source domain to the corresponding output graphs in the target domain via deep learning. Emerging as a new while important problem, deep graph transformation has multiple applications in many areas, such as molecule optimization (Shi et al, 2020;Do et al, 2019) and malware confinement in cyber security (Guo et al, 2019a). Considering the entities that are being transformed during the translation process, there are three categories of sub-problems: node transformation, edge transformation, and node-edge-co-transformation.…”

Interpretable Deep Graph Generation with Node-edge Co-disentanglement

“…Graphite [18] and VGAE [26] encode the nodes of each graph into node-level embeddings and predict the links between each pair of nodes to generate a graph. Some conditional graph generation methods also provide powerful graph encoders and decoders for attributed graphs where both node and edge attributes are considered [19,20].…”

Meta Graph Attention on Heterogeneous Graph with Node-Edge Co-evolution

“…On the one hand, EGNN [5] exploits edge features by considering multi-dimensional edge features as multichannel signals and applies a graph attention operation to each channel separately. [7] formulates a multi-attributed graph translation problem and proposes a multi-block translation architecture called NEC-DGT to tackle this problem, in which the hidden edge and node states are co-evolved. On the other hand, methods like R-GCN [17], HAN [20], HetSANN [11] have been proposed to analyze the heterogeneity in nodes or edges.…”

“…Experimental setup. We compare against two inductive GNN-based baselines: GraphSage [9] and NEC-DGT [7]. GraphSage considers graphs as the homogeneous graph with node attributes.…”