Online Program Home
My Program

Abstract Details

Activity Number: 619 - Spatial and Spatial-Temporal Statistics
Type: Topic Contributed
Date/Time: Thursday, August 2, 2018 : 8:30 AM to 10:20 AM
Sponsor: Section on Statistics and the Environment
Abstract #329791
Title: A Non-Collapsing Particle Filter for a High-Dimensional Cellular Automata Model of Traffic Flow
Author(s): Thomas Bengtsson*
Companies: Genentech
Keywords: Particle FIlter; Forecasting; Prediction; Timer series; Kalman FIlter; High-dimension

Particle filters (PF) used to track dynamical systems from noisy time series observations are known to quickly loose ensemble spread in high-dimensional systems (Snyder at al 2008, Bengtsson et al 2008). For systems of even modest dimension, this lack of variability in the updated (i.e. posterior) ensemble leads to a collapse onto a single particle and consequent filter failure. The construction of sequential importance samplers for high-dimensional systems is therefore inherently challenging. Based on ideas similar to that of the Auxiliary PF (Pitt, 1999), this talk details a sequential importance sampler that does not collapse in high-dimensional model based on a cellular automata scheme to model traffic flow (Helbing, 1998). To understand why the presented high-dimensional PF does not collapse, we explore the effective dimension of the generated and compare the empirical results to theoretical bound set forth in Snyder et al (2015).

Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program