Infinite Mixture Prototypes for Few-Shot Learning

TL;DR

This paper introduces infinite mixture prototypes for few-shot learning, allowing adaptive class representations that improve accuracy and robustness across various data complexities and settings.

Contribution

It presents a novel method that models classes with multiple clusters, interpolating between nearest neighbor and prototypical approaches, enhancing performance in few-shot and semi-supervised learning.

Findings

25% accuracy improvement over prototypical networks on few-shot tasks

Achieves state-of-the-art semi-supervised clustering accuracy

Can perform purely unsupervised clustering with improved flexibility

Abstract

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 inferring the number of clusters, infinite mixture prototypes interpolate between nearest neighbor and prototypical representations, which improves accuracy and robustness in the few-shot regime. We show the importance of adaptive capacity for capturing complex data distributions such as alphabets, with 25% absolute accuracy improvements over prototypical networks, while still maintaining 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.