mapping from the continuous set to the discrete one

[feiran-l]

Hi! Thank you for sharing the nice work. In the paper, you mentioned that the discrete {0, 1} constraint is relaxed to the continuous range [0, 1] for the Frank-Wolfe purpose. Then how do you map the continuous result back to satisfy the binary constraint? E.g., by doing row-maxing or something?

[MhYao2014]

Hi! Thank you for sharing the nice work. In the paper, you mentioned that the discrete {0, 1} constraint is relaxed to the continuous range [0, 1] for the Frank-Wolfe purpose. Then how do you map the continuous result back to satisfy the binary constraint? E.g., by doing row-maxing or something?

I have read the Matlab code. They applied the Hungarian algorithm to \hat{P} (the relaxed continuous permutation matrix) to get the binary version. You can find it in ZAC.m