AUTOENCODER-BASED GRAPH CONSTRUCTION FOR SEMI-SUPERVISED LEARNING

摘要

Disclosed is an autoencoder-based graph construction technique for semi-supervised learning. A graph construction process performed by a graph construction system according to one embodiment may comprise the steps of: generating an input vector by merging a feature vector with a label; by using the generated input vector, simultaneously training a discriminator configured for obtaining the feature vector, and an autoencoder configured for constructing a graph; and constructing a graph on the basis of a prediction result of unlabeled data obtained as a training result of performing the training.