A Class of Spatial Filtering Problems with Unknown Spatial Observations

04/30/2023

∙

We consider a class of high-dimensional spatial filtering problems, where the spatial locations of the observations are unknown and driven by the unobserved signal. This problem is exceptionally challenging as not only is the problem of high-dimensions in the signal, but the model for the signal yields longer-range time dependencies on this object. Motivated by this model we revisit a lesser-known and exact computational methodology from Centanni & Minozzo (2006a) (see also Martin et al. (2013)) designed for filtering of point-processes. We adapt the methodology for this new class of problem. The algorithm is implemented on high-dimensional (of the order of 10^4) rotating shallow water model with real and synthetic observational data from ocean drifters. In comparison to existing methodology, we demonstrate a significant improvement in speed and accuracy.

READ FULL TEXT

A Class of Spatial Filtering Problems with Unknown Spatial Observations

A Lagged Particle Filter for Stable Filtering of certain High-Dimensional State-Space Models

Adaptive Change Point Monitoring for High-Dimensional Data

An Ensemble Score Filter for Tracking High-Dimensional Nonlinear Dynamical Systems

Particle representation for the solution of the filtering problem. Application to the error expansion of filtering discretizations

The One Step Malliavin scheme: new discretization of BSDEs implemented with deep learning regressions

Using topological autoencoders as a filtering function for global and local topology

Representing ill-known parts of a numerical model using a machine learning approach

A Class of Spatial Filtering Problems with Unknown Spatial Observations

Related Research

A Lagged Particle Filter for Stable Filtering of certain High-Dimensional State-Space Models

Adaptive Change Point Monitoring for High-Dimensional Data

An Ensemble Score Filter for Tracking High-Dimensional Nonlinear Dynamical Systems

Particle representation for the solution of the filtering problem. Application to the error expansion of filtering discretizations

The One Step Malliavin scheme: new discretization of BSDEs implemented with deep learning regressions

Using topological autoencoders as a filtering function for global and local topology

Representing ill-known parts of a numerical model using a machine learning approach