search

wyharveychen / CloserLookFewShot

source code to ICLR'19, 'A Closer Look at Few-shot Classification'
Other
1.13k stars 268 forks source link

Could anyone explain why don't use validation for baseline? #61

Open rainylt opened 3 years ago

rainylt commented 3 years ago

If we don't use validation and don't calculate accuracy during training, how can we know when to stop and how it performs in the training set, and whether it has overfitted? Just stare at the loss?

ashok-arjun commented 3 years ago

With baseline, we cannot afford to train a linear classifier (the last layer) everytime we do a validation.

However with protonets, validation is just inference. That's why