search

biomedia-mira / masf

Domain Generalization via Model-Agnostic Learning of Semantic Features
Apache License 2.0
146 stars 18 forks source link

A question about parameters update #12

Closed kjw11 closed 3 years ago

kjw11 commented 4 years ago

Hi, sorry for bothering.

I notiticed that in your method, the D_tr updatas the model parameters at first, then calculate Global loss and Local loss on updated model. But I did't understand what role the update operation is play in your method? From my understanding, the update operation in MAML method is used to simulate the finetuning (adaptation) process when implementing on few-shot learning. Is my understanding wrong?

Thanks for you attention and your time.

carrenD commented 3 years ago

Hi the model update on D_tr is supervised learning on task loss. Thanks!