- Keywords: Generative Flow, Normalizing Flow, Multi-scale Architecture, RealNVP, Dimension Factorization
- TL;DR: Data-dependent factorization of dimensions in a multi-scale architecture based on contribution to the total log-likelihood
- Abstract: Deep generative modeling using flows has gained popularity owing to the tractable exact log-likelihood estimation with efficient training and synthesis process. However, flow models suffer from the challenge of having high dimensional latent space, same in dimension as the input space. An effective solution to the above challenge as proposed by Dinh et al. (2016) is a multi-scale architecture, which is based on iterative early factorization of a part of the total dimensions at regular intervals. Prior works on generative flows involving a multi-scale architecture perform the dimension factorization based on a static masking. We propose a novel multi-scale architecture that performs data dependent factorization to decide which dimensions should pass through more flow layers. To facilitate the same, we introduce a heuristic based on the contribution of each dimension to the total log-likelihood which encodes the importance of the dimensions. Our proposed heuristic is readily obtained as part of the flow training process, enabling versatile implementation of our likelihood contribution based multi-scale architecture for generic flow models. We present such an implementation for the original flow introduced in Dinh et al. (2016), and demonstrate improvements in log-likelihood score and sampling quality on standard image benchmarks. We also conduct ablation studies to compare proposed method with other options for dimension factorization.