The paper introduces the SGVB estimator by reparameterizing continuous approximate-posterior samples as differentiable transformations of auxiliary noise, and introduces AEVB to jointly optimize an approximate inference model and a generative model for i.i.d. data with per-datapoint continuous latent variables. · CiteArk
The paper introduces the SGVB estimator by reparameterizing continuous approximate-posterior samples as differentiable transformations of auxiliary noise, and introduces AEVB to jointly optimize an approximate inference model and a generative model for i.i.d. data with per-datapoint continuous latent variables.
来源：source-paper:PDF pp. 1, 3-5; Abstract, Sections 2.3-2.4, equations (4)-(8), Algorithm 1, and Section 3
