The abstract and slides from Len‘s talk at the second Summer of V’s workshop.
I provide a short tutorial on the particle filter, a simulation-based technique for fitting complex time-series models to data. My motivating example is Bayesian modelling of British grey seal population dynamics. I outline the basic idea of particle filtering, and some tricks to make the algorithm more efficient in different circumstances. Some of these tricks make the algorithm look very Darwinian: simulations that match the data well get to reproduce more and can even mutate, while those that do not die out. Some tricks go beyond what the natural world can manage, e.g., tempering the selection pressure to increase robustness, and taking computationally inexpensive look-aheads to see which simulations will closely match future data. I provide some general guidance as to which problems are well suited to particle filtering approaches.