Tensor-Train Methods for Sequential State and Parameter Estimation in State-Space Models

The next statistics seminar will be presented by Dr Yiran Zhao from School of Mathematics and Statistics.

Title: Tensor-Train Methods for Sequential State and Parameter Estimation in State-Space Models
Speaker: Yiran Zhao
Time and location : 1-2pm on Carslaw 275 or Zoom
Abstract :

Numerous real-world applications require the estimation, forecasting, and control of dynamic systems using incomplete and indirect observations. These problems can be formulated as state-space models, where the challenge lies in learning the model states and parameters from observed data. We present new tensor-based sequential Bayesian learning methods that jointly estimate parameters and states. Our methods provide manageable error analysis and potentially mitigate the particle degeneracy encountered in many particle-based approaches. Besides offering new insights into algorithmic design, our methods naturally incorporate conditional transports, enabling filtering, smoothing, and parameter estimation within a unified framework.