The Infinite Factorial Hidden Markov Model Poster: W63 Jurgen Van Gael, Yee Whye Teh, Zoubin Ghahramani The Markov Indian Buffet Process The Markov Indian Buffet Process (mIBP) defines a distribution over an infinite number of binary Markov chains. The distribution has very similar properties to the IBP but, like the iHMM, introduces Markov correlation between the datapoints. Features ....... The IBP defines a nonparametric factorial distribution for exchangeable datapoints. Infinite Factorial HMM A nonparametric version of the factorial Hidden Markov Model. ® ° am m = 1¢¢¢1 s0m s1m s2m ± H bm µm sT m y1 y2 yT Time ............. The iHMM defines a nonparametric distribution where datapoints are related through a Markov chain. ° Application: blind source separation with learning of # hidden sources. Factorial Model Mixture Model or Factorial HMM ¯ HMM ® H ¼k µk k = 1¢¢¢1 s0 s1 y1 s2 IBP mIBP/iFHMM y2 DP Mixture iHMM