Redatuming physical systems using symmetric autoencoders

TL;DR

This paper introduces SymAE, an autoencoder architecture that leverages symmetry and stochastic regularization to disentangle meaningful signals from nuisances in physical systems, enabling the creation of standardized virtual data instances.

Contribution

It proposes a novel autoencoder framework, SymAE, that uses symmetry and regularization to improve data redatuming in systems with hidden states and measurement nuisances.

Findings

SymAE effectively disentangles coherent and nuisance information.

The method enables uniform virtual data generation across measurements.

SymAE outperforms traditional approaches in data redatuming tasks.

Abstract

This paper considers physical systems described by hidden states and indirectly observed through repeated measurements corrupted by unmodeled nuisance parameters. A network-based representation learns to disentangle the coherent information (relative to the state) from the incoherent nuisance information (relative to the sensing). Instead of physical models, the representation uses symmetry and stochastic regularization to inform an autoencoder architecture called SymAE. It enables redatuming, i.e., creating virtual data instances where the nuisances are uniformized across measurements.