Based On Regularized Robust Semantic Autoencoder For Zero Shot Learning Method

In recent years,benefit from the development of deep learning,image recognition technology has made great achievements in many intelligent recognition fields.Deep learning relies on a amount of labeled samples,which makes it unapplicable in many actual scenarios where the labeled sample size is insufficient.Zero shot learning is a transfer learning method inspired by the human learning mechanism,which makes the computer approach realizing real artificial intelligence,and so that it has considerable research significance and value.In this paper,based on the existing methods,we study the semantic autoencoder of zero shot learning,and propose the regularized semantic autoencoder method and the robust Huber semantic autoencoder method.In the aspect of algorithm,we use the classical Alternating Direction Method of Multipliers(ADMM)algorithm and symmetric Gauss-Seidel based Alternating Direction Method of Multipliers(s GS-ADMM)algorithm to solve the model,and in theory,we give the convergence analysis of the two models.Finally,we conduct a large number of numerical experiments on five small and medium-sized datasets.The experimental results show that under the traditional zero shot learning setting and generalized zero shot learning setting,the classification effect of our two methods are better than the existing semantic semantic autoencoder.