Entropy and Mutual Information
Entropy and Mutual Information
Entropy and Mutual Information
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Convexity <strong>and</strong> Concavity of MII(X; Y ) is a concave function of p(x) given p(y|x)<strong>and</strong> a convex function of p(y|x) given p(x).To prove I(X; Y ) is concave in p(x) given p(y|x),we use I(X; Y ) = H(Y ) − H(Y |X).H(Y ){λp 1 (x) + (1 − λ)p 2 (x)} = H(Y ){λp 1 (y) + (1 − λ)p 2 (y)}≥ λH(Y ){p 1 (y)} + (1 − λ)H(Y ){p 2 (y)}= λH(Y ){p 1 (x)} + (1 − λ)H(Y ){p 2 (x)},So the first term H(Y ) is concave in p(x). Thesecond term H(Y |X) is linear in p(x), which is bothconcave <strong>and</strong> convex. Therefore the rhs is a concavefunctions of p(x).<strong>Entropy</strong> <strong>and</strong> <strong>Mutual</strong> <strong>Information</strong> – p. 24