Enhancing Interpretability in Generative Modeling: Statistically Disentangled Latent Spaces Guided by Generative Factors in Scientific Datasets

TL;DR

This paper introduces Aux-VAE, a novel variational autoencoder architecture that enhances interpretability by statistically disentangling latent variables aligned with physical generative factors in scientific datasets.

Contribution

We propose Aux-VAE, a minimally modified VAE that incorporates auxiliary variables to achieve disentanglement guided by prior statistical knowledge.

Findings

Aux-VAE effectively disentangles latent factors in complex datasets.

The method outperforms existing models in interpretability and accuracy.

Validated on astronomical simulations and other scientific datasets.

Abstract

This study addresses the challenge of statistically extracting generative factors from complex, high-dimensional datasets in unsupervised or semi-supervised settings. We investigate encoder-decoder-based generative models for nonlinear dimensionality reduction, focusing on disentangling low-dimensional latent variables corresponding to independent physical factors. Introducing Aux-VAE, a novel architecture within the classical Variational Autoencoder framework, we achieve disentanglement with minimal modifications to the standard VAE loss function by leveraging prior statistical knowledge through auxiliary variables. These variables guide the shaping of the latent space by aligning latent factors with learned auxiliary variables. We validate the efficacy of Aux-VAE through comparative assessments on multiple datasets, including astronomical simulations.