GMNN | Graph Markov Neural Networks | Machine Learning library

GMNN is a Python library typically used in Artificial Intelligence, Machine Learning, Deep Learning, Pytorch applications. GMNN has no bugs, it has no vulnerabilities and it has low support. However GMNN build file is not available. You can download it from GitHub.

GMNN integrates statistical relational learning methods (e.g., relational Markov networks and Markov logic networks) and graph neural networks (e.g., graph convolutional networks and graph attention networks) for semi-supervised object classification. GMNN uses a conditional random field to define the joint distribution of all the object labels conditioned on object features, and the framework can be optimized with a pseudolikelihood variational EM algorithm, which alternates between an E-step and M-step. In the E-step, we infer the labels of unlabeled objects, and in the M-step, we learn the parameters to maximize the pseudolikelihood. To benefit training such a model, we introduce two graph neural networks in GMNN, i.e., GNNp and GNNq. GNNq is used to improve inference by learning effective object representations through feature propagation. GNNp is used to model local label dependency through local label propagation. The variational EM algorithm for optimizing GMNN is similar to the co-training framework. In the E-step, GNNp annotates unlabeled objects for updating GNNq, and in the M-step, GNNq annotates unlabeled objects for optimizing GNNp. GMNN can also be applied to many other applications, such as unsupervised node representation learning and link classification. In this repo, we provide codes for both semi-supervised object classification and unsupervised node representation learning.