1: \begin{abstract}

2:     We study source localization from high dimensional M/EEG data by extending

3:     a multiscale method based on Entropic inference

4:     devised to increase the spatial resolution of inverse problems .

5:     This method is used to construct informative  

6:     prior distributions in  a manner inspired 

7:     in the  context of fMRI (Amaral et al 2004 \cite{Amaral2004}) .

8:     We construct a  set of renormalized lattices 

9:     that approximate the cortex region where the source activity is located and

10:     address the related problem

11:     of defining the relevant variables in a coarser scale representation of the cortex. The priors 

12:     can be used in conjunction with other Bayesian methods such as the 

13:     Variational Bayes  method (VB, Sato et al 2004 \cite{Sato2004}).

14:     The central point of the algorithm 

15:     is that it uses  a posterior obtained at a coarse scale 

16:     to induce a prior at the next

17:     finer scale stage of the problem. 

18:     We present results which suggest,

19:     on simulated data, that this way of including prior information is a useful aid for the source location problem. This is judged by the rate and magnitude 

20:     of errors in source localization. 

21:     Better convergence times are also achieved. We also present

22:     results on public data collected during a face recognition task. 

23:   \end{abstract}

24: