Here is a useful review on "Multivariate Mutual Information". I'm working on the negative interaction in a tri-variate problem. The definition of semi-independent distribution caught my attraction. From [Han'80: Multiple Mutual Informations and Multiple Interactions in Frequency Data], a tri-variate distribution (U,V,Y) is semi-independent if
Pr{U V Y} = Pr{U} Pr{V Y} + Pr{V} Pr{U Y} + Pr{Y} P r{U V} − 2 Pr{U} Pr{V} Pr{Y}
