Quantitative Understanding of VAE by Interpreting ELBO as Rate Distortion Cost of Transform Coding.

VAE (Variational autoencoder) estimates the posterior parameters (mean and variance) of latent variables corresponding to each input data. While it is used for many tasks, the transparency of the model is still an underlying issue. This paper provides a quantitative understanding of VAE property by interpreting ELBO maximization as Rate-distortion optimization of transform coding. According to the Rate-distortion theory, the optimal transform coding is achieved by using PCA-like orthonormal (orthogonal and unit norm) transform. From this analogy, we show theoretically and experimentally that VAE can be mapped to an implicit orthonormal transform with a scale factor derived from the posterior parameter. As a result, the quantitative importance of each latent variable can be evaluated like the eigenvalue of PCA. We can also estimate the data probabilities in the input space from the prior, loss metrics, and corresponding posterior parameters.