Graph Agreement Models for Semi-Supervised Learning

Otilia Stretcu, Krishnamurthy Viswanathan, Dana Movshovitz-Attias, Anthony Platanios, Sujith Ravi, Andrew Tomkins

Abstract: Graph-based algorithms are one of the most successful paradigms for solving semi-supervised learning problems. Recent work on graph convolutional networks, neural graph learning methods and their variants has successfully combined the expressiveness of neural networks with graph structures. We propose a technique that can be applied to these methods and that achieves new state-of-the-art results on several established semi-supervised learning datasets. Traditional graph-based algorithms, such as label propagation, were designed with the underlying assumption that the label of a node can be imputed from that of the neighboring nodes and the edge weights. However, most real world graphs are either noisy or have edges that do not correspond to label agreement uniformly across the graph. To address this, we propose Graph Agreement Models (GAM), which introduce an auxiliary model that predicts the likelihood of two nodes sharing the same label as a learned function of their features. This agreement model is then used when training a node classification model by encouraging agreement only for those pairs of nodes that it deems likely to have the same label, thus guiding its parameters to better local optima. The classification and agreement models are trained jointly in a co-training fashion. We perform experiments on several established datasets and demonstrate that our method can boost the accuracy of multiple node classification models by up to 20% and obtain state-of-the-art results.

Code Link: https://github.com/tensorflow/neural-structured-learning

CMT Num: 4699