General Invertible Transformations for Flow-based Generative Modeling

TL;DR

This paper introduces a new class of invertible transformations for flow-based generative models, proposing novel coupling layers that improve performance on digit data tasks compared to existing methods.

Contribution

It presents a general framework for invertible transformations and introduces two new coupling layers that enhance flow-based generative modeling.

Findings

New invertible transformations derived from reversible logic.

Proposed coupling layers outperform standard layers in digit data experiments.

Enhanced results in Integer Discrete Flows (IDF) and RealNVP.

Abstract

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.