Self-Supervised Variational Auto-Encoders
This work addresses data compression and generation tasks in AI, offering a flexible model for trading off memory and quality, but it appears incremental as it builds upon existing VAE frameworks.
The authors tackled density estimation, compression, and data generation by introducing self-supervised Variational Auto-Encoders (selfVAE), which use deterministic and discrete variational posteriors to enable conditional and unconditional sampling while simplifying the objective function, and demonstrated performance on benchmark image datasets like Cifar10, Imagenette64, and CelebA.
Density estimation, compression and data generation are crucial tasks in artificial intelligence. Variational Auto-Encoders (VAEs) constitute a single framework to achieve these goals. Here, we present a novel class of generative models, called self-supervised Variational Auto-Encoder (selfVAE), that utilizes deterministic and discrete variational posteriors. This class of models allows to perform both conditional and unconditional sampling, while simplifying the objective function. First, we use a single self-supervised transformation as a latent variable, where a transformation is either downscaling or edge detection. Next, we consider a hierarchical architecture, i.e., multiple transformations, and we show its benefits compared to the VAE. The flexibility of selfVAE in data reconstruction finds a particularly interesting use case in data compression tasks, where we can trade-off memory for better data quality, and vice-versa. We present performance of our approach on three benchmark image data (Cifar10, Imagenette64, and CelebA).