Title: Fully Variational Noise-Contrastive Estimation. Christopher Zach is a research professor in the research group Computer vision and medical image analysis at Chalmers University.

Abstract:
By using the underlying theory of proper scoring rules (which I will briefly introduce), a family of noise-contrastive estimation (NCE) methods can be designed, which are tractable for latent variable models. Both terms in the underlying NCE loss, the one using data samples and the one using noise samples, can be lower-bounded as in variational Bayes, therefore I call this family of losses "fully variational noise-contrastive estimation" (as opposed to an already existing variational noise-contrastive estimation method). Variational autoencoders turn out to be a particular example in this family and therefore can be also understood as separating real data from synthetic samples using an appropriate classification loss. I will further discuss other instances in this family of fully variational NCE objectives and indicate differences in their empirical behavior.