Talker: Daichi Mochihashi (ISM)
Title: Multiresolution Log Gaussian Cox point processes
Abstract: Inferring latent intensities of Poisson point processes is
of fundamental importance in many fields such as seismology,
ecology, neuroscience, and activity recognition, to name a few.
However, so far only quite simple models have been introduced for
inferring the intensity, such as homogeneous Log Gaussian Cox
processes (Moller+, 1998). In this study, we introduce
multiresolution Gaussian process (Fox and Dunson 2012) as a latent
process governing observed point process. This is a recursive
change-point problem only from discrete observations, and we
estimate it with interleaved Elliptical slice sampling of
latent Gaussian processes. We show some experiments on synthetic
data, neural firing data and ecological data on forestry.