?url_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Adc&rft.title=The+Role+of+Global+Labels+in+Few-Shot+Classification+and+How+to+Infer+Them&rft.creator=Wang%2C+R&rft.creator=Pontil%2C+M&rft.creator=Ciliberto%2C+C&rft.description=Few-shot+learning+is+a+central+problem+in+meta-learning%2C+where+learners+must+quickly+adapt+to+new+tasks+given+limited+training+data.+Recently%2C+feature+pre-training+has+become+a+ubiquitous+component+in+state-of-the-art+meta-learning+methods+and+is+shown+to+provide+significant+performance+improvement.+However%2C+there+is+limited+theoretical+understanding+of+the+connection+between+pre-training+and+meta-learning.+Further%2C+pre-training+requires+global+labels+shared+across+tasks%2C+which+may+be+unavailable+in+practice.+In+this+paper%2C+we+show+why+exploiting+pre-training+is+theoretically+advantageous+for+meta-learning%2C+and+in+particular+the+critical+role+of+global+labels.+This+motivates+us+to+propose+Meta+Label+Learning+(MeLa)%2C+a+novel+meta-learning+framework+that+automatically+infers+global+labels+to+obtains+robust+few-shot+models.+Empirically%2C+we+demonstrate+that+MeLa+is+competitive+with+existing+methods+and+provide+extensive+ablation+experiments+to+highlight+its+key+properties.&rft.publisher=NeurIPS&rft.date=2021&rft.type=Proceedings+paper&rft.language=eng&rft.source=+++++In%3A++Proceedings+of+the+35th+Conference+on+Neural+Information+Processing+Systems+(NeurIPS+2021).++(pp.+pp.+27160-27170).++NeurIPS+(2021)+++++&rft.format=text&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10151383%2F1%2Fthe_role_of_global_labels_in_f.pdf&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10151383%2F&rft.rights=open