Identifiability and prediction in ecological interaction network models

Abstract: 

Many processes in ecology are fundamentally driven by known interactions between individuals. Meanwhile the pressing questions, such as species extinction, are naturally posed at the system level. Mapping a microscopic agent-based description to a mesoscopic aggregate model is of course a highly non-trivial matter and generally introduces noise as a result of observing only certain selected degrees of freedom. One approach is to frame the mesoscopic dynamics as a continuous-time Markov chain whose states are accounts of taxa abundances and whose transitions are interaction events. Such 'Markovian interaction networks' have previously risen to fame in systems biology as biochemical reaction networks and have likewise been a subject of theoretical ecology for over half a century. However they are not as widely adopted in experimental ecology studies as their infinite-population ODE counterparts. After reviewing the basics of Markovian interaction networks I will focus on their prediction theory which involves stochastic filtering and give a glimpse into an ongoing software project of developing a general solution to integrate Markovian interaction networks into experimental design protocols and facilitate the discourse between theorists and experimentalists around that subject.

zoom link: https://us06web.zoom.us/j/85744990492?pwd=U3l3YUh6ZmNVUjNVOVV4dldxdHdVQT09