The Theoretical Breakthrough of Self-Supervised Learning : Variational Autoencoders and Its Application In Big Data Analysis

Abstract

In the study of big data statistics, a large number of unknown latent variables bring great difficulties to modeling. Variational autoencoders(VAE) can overcome the shortcomings of traditional variational methods such as low efficiency and poor generality, and provide an efficient and extensible framework for variational posteriori inference and approximate maximum likelihood learning based on gradient. On the basis of reviewing the development history of variational autoencoders and taking the deep latent variable model(DLVM) as an example, this paper introduces the basic principle of variational autoencoders and analyzes its application under the background of big data. The problems in theory and application of variational autoencoders are presented, as well as the topics to be further studied. The combination of variational autoencoders and other statistical modeling methods may become a new idea for big data research.