Mixture variational autoencoders

Shuoran Jiang, Yarui Chen, Jucheng Yang, Chuanlei Zhang, Tingting Zhao

Abstract: Highlights•A mixture variational autoencoders (MVAES) is proposed.•MVAES use a discrete latent variable to indicate the components of variational autoencoders.•Neural networks are introduced to approximate the variational posteriors of latent variables.•The Stick-Breaking parameterization and reparameterization trick are proposed to optimize the variational lower bound.•Stochastic gradient variational Bayes estimator is used to calculate the variational lower bound of MVAEs.