DE-VAE: Revealing Uncertainty in Parametric and Inverse Projections with Variational Autoencoders using Differential Entropy

TL;DR

DE-VAE introduces an uncertainty-aware variational autoencoder that leverages differential entropy to improve parametric and invertible projections of multidimensional data, effectively handling out-of-distribution samples and enabling uncertainty analysis.

Contribution

It presents DE-VAE, a novel variational autoencoder that incorporates differential entropy to enhance projection accuracy and uncertainty quantification in data embeddings.

Findings

DE-VAE achieves comparable accuracy to existing AE-based methods.

It enables analysis of embedding uncertainty.

Demonstrated effectiveness on four benchmark datasets.

Abstract

Recently, autoencoders (AEs) have gained interest for creating parametric and invertible projections of multidimensional data. Parametric projections make it possible to embed new, unseen samples without recalculating the entire projection, while invertible projections allow the synthesis of new data instances. However, existing methods perform poorly when dealing with out-of-distribution samples in either the data or embedding space. Thus, we propose DE-VAE, an uncertainty-aware variational AE using differential entropy (DE) to improve the learned parametric and invertible projections. Given a fixed projection, we train DE-VAE to learn a mapping into 2D space and an inverse mapping back to the original space. We conduct quantitative and qualitative evaluations on four well-known datasets, using UMAP and t-SNE as baseline projection methods. Our findings show that DE-VAE can create parametric and inverse projections with comparable accuracy to other current AE-based approaches while enabling the analysis of embedding uncertainty.