Coarse Grained Exponential Variational Autoencoders
This addresses a bottleneck in variational learning for researchers in machine learning, offering a method to use more sophisticated posterior distributions, but it appears incremental as it builds on existing VAE frameworks.
The paper tackled the limitation of variational autoencoders (VAEs) using simplified distributions like Gaussian or categorical, which mismatch the true posterior, by introducing a semi-continuous latent representation that approximates continuous densities with precision and applies polynomial exponential family distributions for inference. The result showed consistent improvements over common VAE models, though no specific numbers were provided.
Variational autoencoders (VAE) often use Gaussian or category distribution to model the inference process. This puts a limit on variational learning because this simplified assumption does not match the true posterior distribution, which is usually much more sophisticated. To break this limitation and apply arbitrary parametric distribution during inference, this paper derives a \emph{semi-continuous} latent representation, which approximates a continuous density up to a prescribed precision, and is much easier to analyze than its continuous counterpart because it is fundamentally discrete. We showcase the proposition by applying polynomial exponential family distributions as the posterior, which are universal probability density function generators. Our experimental results show consistent improvements over commonly used VAE models.