search

yaoyao-liu / meta-transfer-learning

TensorFlow and PyTorch implementation of "Meta-Transfer Learning for Few-Shot Learning" (CVPR2019)
https://lyy.mpi-inf.mpg.de/mtl/
MIT License
731 stars 147 forks source link

Question about meta-validation during meta-train #14

Open vickichen97 opened 4 years ago

vickichen97 commented 4 years ago

Whether the meta-validation data classes are consistent with the meta-train data classes in one task? For meta-validation during meta-train, the meta_lr is 0 which means ss_weights and fc_weights will not be updated. Whether the updates in fast_fc_weights will not influence fc_weights when the meta_lr is 0 ? Looking forward to your reply,thanks!

yaoyao-liu commented 4 years ago

Answer to Q1: Meta-train, meta-val, and meta-test set share no common class, i.e., the class occurs in meta-train won't occur in meta-val and meta-test.

Answer to Q2: During base-learning, the temporary fc weights are detached from the parameter list so the updates won't be stored. https://github.com/yaoyao-liu/meta-transfer-learning/blob/2fd01a254d580909e5c1c0100347ab71216aad62/pytorch/models/mtl.py#L98

vickichen97 commented 4 years ago

Thank you very much！