César Ojeda (Fraunhofer Institute for Intelligent Analysis and Information Systems IAIS)
We propose a Hawkes point process model for point processes where the excitations are modulated via a Gaussian process prior with a sigmoid link function. This formulation allows for a rich and flexible definition of the correlation aspects of the Hawkes self excitations. Efficient approximate Bayesian inference is achieved via latent variables over the branching structure and data augmentation with P´olya Gamma random variables. We apply our methodology to neural activity from calcium fluorescence recordings, order book data and block creation from the Bitcoin’s blockchain transactions network.
Invited by Manfred Opper
