The Lifebelt Particle Filter: a novel SMC scheme

Speaker's Name: Alice Corbella
Speaker's Affiliation: University of Warwick
Speaker's Research Theme(s): Statistics and Probability
Abstract:
Sequential Monte Carlo (SMC) methods can be applied to discrete State-Space Models (SSMs) on bounded domains, to sample from and marginalise over unknown random variables. Similarly to continuous settings, problems such as particle degradation can arise: proposed particles can be incompatible with the data, lying in low probability regions or outside the boundary constraints, and the discrete system could result in all particles having weights of zero. In this talk, I will present the Lifebelt Particle Filter (LBPF), a novel SMC method for robust likelihood estimation in low-valued count problems. The LBPF combines a standard particle filter with one (or more) lifebelt particles, which, by construction, lie within the boundaries of the discrete random variables, and therefore are compatible with the data. The main benefit of the LBPF is that only one or a few, wisely chosen, particles are sufficient to prevent particle collapse. In the talk, I will also present examples of the use of the LBPF, including parameter inference for a death-and-recovery model of hospitalised patients during an epidemic and state inference for a stochastic SIR epidemic model. I will conclude by outlining recent explorative work around choices of resampling mechanisms within this algorithm and their effect on the LBPF performance. This is joint work with Simon Spencer and Richard Haughey. Corbella A, McKinley TJ, Birrell PJ, Presanis AM, De Angelis D, Roberts GO, Spencer SE. The lifebelt particle filter for robust estimation from low-valued count data. Foundations of Data Science. 2024. https://arxiv.org/abs/2212.04400.

Venue: A17