Table 3.

One-shot accuracy on Omniglot standard split

Model	5-way	10-way	15-way	20-way
Human performance (Lake et al., 2015)	-	-	-	95.5
Pixel kNN (Lake et al., 2013)	-	-	-	21.7
Affine model (Lake et al., 2013)	-	-	-	81.8
Deep Boltzmann Machines (Lake et al., 2013)	-	-	-	62.0
Hierarchial Bayesian Program Learning (Lake et al., 2015)	-	-	-	96.7
Siamese Net (Koch, 2015)	-	-	-	92.0
MetaNet	98.45	97.32	96.4	95.92