Thanks for your impressive paper for a entropy coding perspective in CL !
I am very curious about the gap between the transformation from formular (2) to formular (3) where using Taylor series expansion.
I think the Logmatrix fuction can be expanded, only when the inner matrix can be Diagonalized (such as a symmetric matrix like Z@Z.T). However, the Z1 @ Z2.T in formular (3) seems not.
By the way, the formular (25) will be slightly changed from sum of cij^2 to cij*cji if Z1 @ Z2.T is not symmetric.
Therefore, does that means the Taylor series expansion used in entropy coding fuction MEC with Z1,Z2 is only a numerical approximation method?
Thanks for your impressive paper for a entropy coding perspective in CL ! I am very curious about the gap between the transformation from formular (2) to formular (3) where using Taylor series expansion.
I think the Logmatrix fuction can be expanded, only when the inner matrix can be Diagonalized (such as a symmetric matrix like Z@Z.T). However, the Z1 @ Z2.T in formular (3) seems not. By the way, the formular (25) will be slightly changed from sum of cij^2 to cij*cji if Z1 @ Z2.T is not symmetric.
Therefore, does that means the Taylor series expansion used in entropy coding fuction MEC with Z1,Z2 is only a numerical approximation method?