China
Africa
Explore chapters and articles related to this topic
Networked Evolutionary Games
Published in Haitao Li, Guodong Zhao, Peilian Guo, Zhenbin Liu, Analysis and Control of Finite-Value Systems, 2018
Haitao Li, Guodong Zhao, Peilian Guo, Zhenbin Liu
Directed graph: If the FNG is not symmetric, the directed edge is used to distinguish different roles of two players. Assume (i,j)∈E $ (i, j) \in E $ , i.e., there is an edge from i to j, then in the game i is player 1 and j is player 2.Homogeneous graph: If the graph is undirected and each node has same degree, or the graph is directed and each node has same in-degree and same out-degree.
Towards knowledge graph reasoning for supply chain risk management using graph neural networks
Published in International Journal of Production Research, 2022
Edward Elson Kosasih, Fabrizio Margaroli, Simone Gelli, Ajmal Aziz, Nick Wildgoose, Alexandra Brintrup
As shown in Figures 5 and 8, the model can predict multiple types of hidden relationships, while SNLP and GNN-LP can only predict supplies_to edge type. This is due to the fundamental difference in both how we represent the data as a Knowledge Graph instead of a homogeneous graph, and due to the proposed GNN model, which can predict the existence of multiple edge types instead of just one.
An Attention-Based Adversarial Disentangle Heterogeneous Embedding for Improving Node Classification
Published in Cybernetics and Systems, 2022
In recent years, due to aforementioned challenges of multiple meta-path-based and fusion approach in HIN embedding domain, several researchers have dedicated their studies (B. Hu, Fang, and Shi 2019; R. Wang et al. 2021) in the HIN disentangled embedding approach. This approach is mainly inspired from the disentangled data representation paradigm (Kim and Mnih 2018; Ma et al. 2019) which are commonly studied in CV area. There are remarkable disentanglement representation learning studies in information network have been introduced very recently, including the factorize-able GNN (Y. Yang, Feng, et al. 2020) and IPGDN (Liu et al. 2019) frameworks which have concentrated on extracting the disentangled latent features from information network that are influenced by relationships between node pairs. However, these works are mainly focused on dealing with homogeneous graph-structured data representation problem only, thus they might be unable to effectively cope with the semantic-varied representation learning task in network heterogeneity context. By largely ignoring the heterogeneity of the given network, the previous network disentanglement embedding techniques can’t cover the goal of extracting disentangled intrinsic and semantic-specific latent features from semantic-rich heterogeneous networks. To deal with the HIN-based disentangled representation learning problem, recently R. Wang et al. have proposed the HEAD model (R. Wang et al. 2021) which enable to extract the disentangled and unambiguous meta-path-based node representations from heterogeneous networks by applying the adversarial graph embedding paradigm within the VAE architecture. This approach is mainly inspired from previous adversarial graph learning technique of the HeGAN model (B. Hu, Fang, and Shi 2019). Through the adversarial node embedding generative process in HEAD model, the learnt and reconstructed meta-path-based node embeddings can be efficiently purified. Moreover, this approach also supports to enrich the graph-structural/semantic-specific information of learnt node embeddings as well as significantly reduced noise inferences during the task-driven fine-tuning process. However, the recent graph disentangled embedding techniques also suffered few limitations related to the capability of efficient fused semantic-varied embedding interpretation, especially in the context of multiple meta-paths are used, and latent feature alignment for transforming heterogeneous node representations into unified and better task-driven-friendly embedding space. Due to the remained challenges of heterogeneous network disentangled embedding approach, there is still room for further studies and enhancements in this research direction.