Infinite Mixture Prototypes for Few-shot Learning

We propose infinite mixture prototypes to adaptively represent both simple and complex data distributions for few-shot learning. Our infinite mixture prototypes represent each class by a set of clusters, unlike existing prototypical methods that represent each class by a single cluster. By infer-ring the number of clusters, infinite mixture prototypes interpolate between nearest neighbor and prototypical representations, which improves ac-curacy and robustness in the few-shot regime. We show the importance of adaptive capacity for capturing complex data distributions such as alpha-bets, with 25% absolute accuracy improvements over prototypical networks, while still maintain-ing or improving accuracy on the standard Omniglot and mini-ImageNet benchmarks. In clustering labeled and unlabeled data by the same clustering rule, infinite mixture prototypes achieves state-of-the-art semi-supervised accuracy. As a further capability, we show that infinite mixture prototypes can perform purely unsupervised clustering, unlike existing prototypical methods.