General Invertible Transformations for Flow-based Generative Modeling
This work provides new building blocks for researchers and practitioners working with flow-based generative models, potentially leading to improved performance in generative tasks.
This paper introduces a new class of invertible transformations and two new coupling layers for flow-based generative models. When applied to Integer Discrete Flows (IDF) on digit data, these new coupling layers achieve better results than standard coupling layers used in IDF and RealNVP.
In this paper, we present a new class of invertible transformations with an application to flow-based generative models. We indicate that many well-known invertible transformations in reversible logic and reversible neural networks could be derived from our proposition. Next, we propose two new coupling layers that are important building blocks of flow-based generative models. In the experiments on digit data, we present how these new coupling layers could be used in Integer Discrete Flows (IDF), and that they achieve better results than standard coupling layers used in IDF and RealNVP.