Conditional Sampling of Variational Autoencoders via Iterated Approximate Ancestral Sampling
This work addresses a computational bottleneck for applications like missing data imputation in VAEs, but it is incremental as it builds on existing sampling methods.
The paper tackled the problem of conditional sampling in variational autoencoders (VAEs), which is computationally intractable, by addressing limitations of the Metropolis-within-Gibbs sampler that gets stuck due to structured latent spaces, and demonstrated improved performance on sampling tasks.
Conditional sampling of variational autoencoders (VAEs) is needed in various applications, such as missing data imputation, but is computationally intractable. A principled choice for asymptotically exact conditional sampling is Metropolis-within-Gibbs (MWG). However, we observe that the tendency of VAEs to learn a structured latent space, a commonly desired property, can cause the MWG sampler to get "stuck" far from the target distribution. This paper mitigates the limitations of MWG: we systematically outline the pitfalls in the context of VAEs, propose two original methods that address these pitfalls, and demonstrate an improved performance of the proposed methods on a set of sampling tasks.