Variational auto-encoders with Student's t-prior
This work addresses the need for more robust data approximation in VAEs, but it is incremental as it modifies an existing method with a different prior distribution.
The authors tackled the problem of improving variational auto-encoders (VAEs) by proposing a new prior structure using the multivariate Student's t-distribution, which resulted in better image reconstruction on Fashion-MNIST data compared to standard Gaussian priors.
We propose a new structure for the variational auto-encoders (VAEs) prior, with the weakly informative multivariate Student's t-distribution. In the proposed model all distribution parameters are trained, thereby allowing for a more robust approximation of the underlying data distribution. We used Fashion-MNIST data in two experiments to compare the proposed VAEs with the standard Gaussian priors. Both experiments showed a better reconstruction of the images with VAEs using Student's t-prior distribution.