Hi,
I guess 2.2 the equation (1) doesn't hold. It should be H(x_i | y) instead of H(\pi(x_i) | y)? Then the equation would make sense.
Also if we further consider the problem, the minimization of entropy H(p) is only dependent on the Bayesian network structure. If as stated by equation (1), then the minimization of entropy H(p) is related not only to the mutual information but also the entropy of H(\pi(x_i) | y). Then the statement of 2.3 and 2.4 will not hold...
So I guess it's a typo?
Please correct me if I'm wrong.
Best Regards,
Xiaolong Shen