1: \begin{abstract}

2: %EEG signals are recorded from multiple subjects.

3: %The signal has inter-subject variability and intra-subject variability.

4: %And number of latent parameter has to be set before learning 

5: %which is time consuming 

6: %and we do not know whether the found value is the best one.

7: %There are related works in the literature 

8: %where the model includes intra/inter  subject variability

9: %with hierarchical structure, 

10: %or probabilistic model is applied to 

11: %NMF with non-parametric setting, 

12: %but not both at the same time.

13: %We propose the hierarchical infinite NMF model

14: %that can model hierarchical structure and infinite model as well.

15: %The proposed model is evaluated through the synthetic data 

16: %and real world data.

17: 

18: %Group analysis of EEG data provides 

19: %information of the commonly occurring patterns for a specific task

20: %and capability of classification as well.

21: We propose a generative model of a group EEG analysis, 

22: based on appropriate kernel assumptions on EEG data.

23: We derive the variational inference update rule 

24: using various approximation techniques.

25: The proposed model outperforms 

26: the current state-of-the-art algorithms

27: in terms of common pattern extraction.

28: The validity of the proposed model is tested on

29: the BCI competition dataset.

30: 

31: \end{abstract}

32: