Code for the paper:
Yinjun Wu, Adam Stein, Jacob Gardner, Mayur Naik, "Learning to Select Pivotal Samples for Meta Re-weighting " (AAAI 2023 oral)
pip install -r requirements.txt
Minimal code block for using our algorithm for selecting meta samples for meta reweighting algorithm:
#select valid samples with rbc
valid_ids, new_valid_representations = get_representative_valid_ids_rbc(train_dataset, criterion, optimizer, trainloader, args, net, valid_count)
#select valid samples with gbc
valid_ids, new_valid_representations = get_representative_valid_ids_gbc(train_dataset, criterion, optimizer, trainloader, args, net, valid_count)
See the jupyter notebook main_example.ipynb for details