Prototype learning with margin-based conditional log-likelihood loss | IEEE Conference Publication | IEEE Xplore