Redatuming physical systems using symmetric autoencoders

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.

Automated summary

This paper examines physical systems with hidden states that are indirectly observed through repeated measurements corrupted by unaccounted-for nuisance parameters. A network-based representation learns to disentangle coherent information relative to the state from incoherent nuisance information relative to the sensing. Instead of using physical models, the representation utilizes symmetry and stochastic regularization to guide an autoencoder architecture called SymAE. It enables the creation of virtual data instances with uniformized nuisances across measurements, a process known as redatuming.