Uncertainty Autoencoders: Learning Compressed Representations via Variational Information Maximization

TL;DR

Uncertainty Autoencoders introduce a novel unsupervised learning framework inspired by compressed sensing, optimizing mutual information to improve high-dimensional data representation and recovery.

Contribution

The paper presents a new autoencoder-based framework that learns both data acquisition and recovery processes through variational mutual information maximization.

Findings

Achieved a 32% average improvement over competing methods in statistical compressed sensing.

Unified approach connecting dimensionality reduction, compressed sensing, and generative modeling.

Demonstrated effectiveness on high-dimensional datasets.

Abstract

Compressed sensing techniques enable efficient acquisition and recovery of sparse, high-dimensional data signals via low-dimensional projections. In this work, we propose Uncertainty Autoencoders, a learning framework for unsupervised representation learning inspired by compressed sensing. We treat the low-dimensional projections as noisy latent representations of an autoencoder and directly learn both the acquisition (i.e., encoding) and amortized recovery (i.e., decoding) procedures. Our learning objective optimizes for a tractable variational lower bound to the mutual information between the datapoints and the latent representations. We show how our framework provides a unified treatment to several lines of research in dimensionality reduction, compressed sensing, and generative modeling. Empirically, we demonstrate a 32% improvement on average over competing approaches for the task of statistical compressed sensing of high-dimensional datasets.