Infinite Markov Models and Related Topics / Unknown Targets and Patterns of Dynamics
By Daichi Mochihashi
In the first part of this talk, I will introduce the Infinite Markov Model, a
nonparametric Bayesian Markov model that allows us to infer latent Markov orders
purely from the observations. [I will also include slides not discussed during
my talk at NIPS 2007 and some research directions.]
The second part of this talk will describe our group's latest work (in submission to CVPR 2008) about time series inference. This model infers a time-varying number of moving target trajectories and patterns of their dynamics simultaneously. Patterns of dynamics are modeled through a Dirichlet process mixtures of Kalman filters, and the inference is performed through Particle filters.