search

DaikiTanaka-UT / JointOptimization

Joint Optimization Framework for Learning with Noisy Labels
45 stars 9 forks source link

About loss function #1

Open ruirui88 opened 5 years ago

ruirui88 commented 5 years ago

Hi, you wrote that your loss function use the Kullback-Leibler (KL)-divergence for the classification loss in the paper. But you impement the cross entropy function not KL in your source code. So which one ？

DaikiTanaka-UT commented 4 years ago

You are right, but the derivative of the KL-div is equal to that of the cross entropy loss. Therefore, the performance is not affected.