Decafs: Disentangled Conditional adversarial Flows

2026-07-21

AI Summary

Although flow-based models excel in generative tasks, their latent spaces often suffer from entangled factors, hindering controllable and interpretable generation. This work proposes a novel conditional generator by introducing Lie group structure into conditional flow models for the first time. The approach aligns a disentangled auxiliary latent space with the original flow latent space via an adversarial loss, preserving model invertibility and simplicity without increasing the flow’s dimensionality. This method effectively achieves latent space disentanglement and enables interpretable conditional generation, outperforming StyleGAN on MNIST and dSprites image generation tasks and achieving state-of-the-art performance on molecular generation benchmarks including QM9, ZINC, and MOSES.

Abstract

Flow-based models have established state-of-the-art performance in generative modeling across domains, but are hard to interpret due to their complex latent embeddings. In particular, the entanglement of generative factors in the latent space hinders controlled generation. We circumvent this issue by appealing to a novel conditional generator based on Lie groups that disentangles an alternative latent space, which is aligned closely with the latent flow space using an adversarial loss. Our approach facilitates interpretable conditional generation while obviating the need to expand the dimensionality of the flow space (owing to its invertibility requirements). The proposed model demonstrates strong performance across conditional image (including, outperforming StyleGAN on MNIST, dSprites) and molecule (using standard QM9, ZINC and MOSES) generation tasks

Problem

• flow-based models
• latent space entanglement
• controlled generation
• interpretability
• conditional generation

Innovation

• disentangled representation
• conditional generation
• flow-based models
• Lie groups
• adversarial learning