function g = nllicaGrad(w, X, beta);
%
% gradient of the negative log likelihood of ICA model 
% using a Generalized Gaussian distribution with parameter beta in 2-d.
% input 
%       w: vectorized form of unmixing matrix W
%       X: data matrix
%       beta: parameter of Generalized Gaussian distribution 
% output
%       g: gradient vector

W = reshape(w,2,2);
N = size(X,2);
WI = 1/(W(1,1)*W(2,2) - W(1,2)*W(2,1))*[ W(2,2) -W(1,2); -W(2,1) ...
                    W(1,1) ];

if beta == 1
  g = N*WI' - sign(W*X)*X';
else
  g = N*WI' - (beta*sign(W*X).*abs(W*X).^(beta-1))*X';
end


g = -g(:)';