Skip to content

Experiments and derivations for ACC2022 paper on variational message passing for online NARMAX identification.

License

Notifications You must be signed in to change notification settings

biaslab/ACC2022-vmpNARMAX

Repository files navigation

ACC2022 VMP-NARMAX

Experiments and supplementary derivations for the paper entitled

"Variational message passing for online polynomial NARMAX identification"

published at the 2022 American Control Conference.

The goal is infer parameters in a polynomial NARMAX model (see ForneyLab node code) and simulate outputs. Typically, (recursive) least-squares or another form of maximum-likelihood estimation is used. In this project, we employ variational Free Energy Minimisation in the form of variational message passing on a Forney-style factor graph.

We run a series of verification experiments on data generated from a NARMAX system, comparing performance as a function of sample size, polynomial order and simulation noise.

Comments

Questions, comments and general feedback can be directed to the issues tracker.

About

Experiments and derivations for ACC2022 paper on variational message passing for online NARMAX identification.

Topics

Resources

License

Stars

Watchers

Forks

Languages