TzK Flow - Conditional Generative Model
This work addresses the challenge of stable and expressive generative modeling with conditional information, offering a method that combines benefits from GANs, VAEs, and flow-based models, but it appears incremental as it builds on existing flow-based and conditional techniques.
The paper tackles the problem of learning tight conditional priors for data distributions by introducing TzK, a conditional probability flow-based model that leverages attributes like style or class, achieving results comparable to state-of-the-art on MNIST and Omniglot datasets.
We introduce TzK (pronounced "task"), a conditional probability flow-based model that exploits attributes (e.g., style, class membership, or other side information) in order to learn tight conditional prior around manifolds of the target observations. The model is trained via approximated ML, and offers efficient approximation of arbitrary data sample distributions (similar to GAN and flow-based ML), and stable training (similar to VAE and ML), while avoiding variational approximations. TzK exploits meta-data to facilitate a bottleneck, similar to autoencoders, thereby producing a low-dimensional representation. Unlike autoencoders, the bottleneck does not limit model expressiveness, similar to flow-based ML. Supervised, unsupervised, and semi-supervised learning are supported by replacing missing observations with samples from learned priors. We demonstrate TzK by training jointly on MNIST and Omniglot datasets with minimal preprocessing, and weak supervision, with results comparable to state-of-the-art.