1 paper
Kai Wang, Xialei Liu, Andy Bagdanov +3
Most meta-learning approaches assume the existence of a very large set of labeled data available for episodic meta-learning of base knowledge. This contrasts with the more realisti…