[Submitted on 9 May 2024]
View a PDF of the paper titled Outlier-robust Kalman Filtering via Generalised Bayes, by Gerardo Duran-Martin and seven different authors
Summary:We derive a novel, provably sturdy, and closed-form Bayesian replace rule for on-line filtering in state-space fashions within the presence of outliers and misspecified measurement fashions. Our technique combines generalised Bayesian inference with filtering strategies such because the prolonged and ensemble Kalman filter. We use the previous to indicate robustness and the latter to make sure computational effectivity within the case of nonlinear fashions. Our technique matches or outperforms different sturdy filtering strategies (reminiscent of these based mostly on variational Bayes) at a a lot decrease computational value. We present this empirically on a variety of filtering issues with outlier measurements, reminiscent of object monitoring, state estimation in high-dimensional chaotic programs, and on-line studying of neural networks.
Submission historical past
From: Gerardo Duran-Martin [view email]
[v1]
Thu, 9 Could 2024 09:40:56 UTC (6,951 KB)