A Factorial Mixture Prior for Compositional Deep Generative Models
This work addresses the challenge of adding structured compositionality to generative models for domains like image analysis, though it appears incremental as it builds on existing deep generative frameworks.
The paper tackles the problem of modeling high-dimensional data as compositional expressions of latent properties by proposing a factorial mixture prior for deep generative models, resulting in a method that can infer discrete properties in unsupervised or semi-supervised settings with empirical evaluation.
We assume that a high-dimensional datum, like an image, is a compositional expression of a set of properties, with a complicated non-linear relationship between the datum and its properties. This paper proposes a factorial mixture prior for capturing latent properties, thereby adding structured compositionality to deep generative models. The prior treats a latent vector as belonging to Cartesian product of subspaces, each of which is quantized separately with a Gaussian mixture model. Some mixture components can be set to represent properties as observed random variables whenever labeled properties are present. Through a combination of stochastic variational inference and gradient descent, a method for learning how to infer discrete properties in an unsupervised or semi-supervised way is outlined and empirically evaluated.