search

easezyc / deep-transfer-learning

A collection of implementations of deep domain adaptation algorithms
MIT License
889 stars 205 forks source link

Domain-specific Classifier Alignment Loss #3

Closed deep0learning closed 5 years ago

deep0learning commented 5 years ago

In your paper, you have mentioned the discrepancy loss for aligning the classifiers. In your code I did not find the discrepancy loss for classifiers. Did I miss that? Can you please explain it?

Thank you in advanced.

easezyc commented 5 years ago

In my code, the loss named l1_loss is the discrepancy loss.

deep0learning commented 5 years ago

Can you explain why we need this loss as all the domain have the same number of classes, i.e., 31 for office-31 dataset ?

easezyc commented 5 years ago

The classifiers are trained on different source domains, hence they might have the disagreement on the prediction for target samples especially the target samples near class boundaries. Intuitively, the same target sample predicted by different classifiers should get the same prediction. (It is descripted in the paper.)