A0841
Title: Autocompound random measures
Authors: Riccardo Corradin - University of Milano-Bioccca (Italy) [presenting]
Fabrizio Leisen - Kings College London (United Kingdom)
Abstract: The aim is to introduce a class of time-dependent nonparametric models, suited for scenarios involving populations observed at distinct discrete times. Starting with an ancestral random measure, it is possible to define a sequence of time-specific random measures by acting on their intensity functions, in the spirit of compound random measures. The resulting family of models exhibits desirable properties, including mathematical tractability, simple expressions for its main summaries, and a closed-form representation of the joint posterior distribution at distinct observed times. Such a model can then be normalized and used as a building block for dynamic population studies, defining tractable species sampling models that evolve over time, or convoluted with a kernel function to obtain time-dependent mixture models.
