search

Optimization-AI / LibAUC

LibAUC: A Deep Learning Library for X-Risk Optimization
https://libauc.org/
MIT License
273 stars 37 forks source link

关于09示例的实验 #15

Closed yanruiD closed 2 years ago

yanruiD commented 2 years ago

对于09示例，论文里消融实验考虑了S1≠S2，代码里是如何体现的，感谢指出，谢谢大佬。

yzhuoning commented 2 years ago

CompositionalLoss是一个dynamic loss，比如iteration0和iteration1的backend分别调用了CELoss和AUCMLoss，间接的实现了S1≠S2，同时也避免了使用多个dataloader。