Browse Teach Curate

Fast Latent-Factor Inversion for High-Dimensional Noisy Dynamical Systems

A scalable latent factor model that infers the hidden state of high-dimensional dynamical systems from noisy measurements. By constraining the factor loading matrix to be orthogonal, it avoids inverting the posterior covariance in the Kalman filter and admits closed-form expectation-maximization parameter updates, giving large exact speedups. It is motivated by and applied to inverse estimation of slow slip events from continuous-GPS geodetic data in the Cascadia region.

Chappie Text

Fast data inversion for high-dimensional dynamical systems from noisy measurements Yizi Lin, ∗

This paper develops a scalable latent factor model for inferring the hidden state of high-dimensional dynamical systems from noisy measurements. Each latent factor process has its own correlation and…