Event Details

Markov-switching discrete-time Hawkes process

Hawkes processes are used in many fields of application (earth sciences, biology, ecology) to describe the occurrence of punctual events with a self-exciting (or auto-inhibiting) character. More precisely, the Hawkes process is a point process whose past-conditional intensity depends on the dates of past events through a so-called kernel function.

We focus here on the case where the basic intensity of the process varies over time according to a Markovian regime. We show that, in the case of an exponential kernel, the discretized version of this process corresponds to a discrete-time hidden Markov model, whose inference can be carried out using standard methods.

We will illustrate the interest of this model for detecting changes in animal behavior from recordings of bat cries.

-- Joint work with Anna Bonnet (LPSM, Sorbonne Université) --