Interpretable Concept-based Prototypical Networks for Few-Shot Learning

TL;DR

This paper introduces an interpretable few-shot learning method that uses human-understandable concepts and concept-specific metric spaces to classify new samples without needing concept annotations, matching state-of-the-art accuracy.

Contribution

It presents a novel concept-based approach for few-shot learning that enhances interpretability while maintaining competitive performance.

Findings

Achieved comparable results to state-of-the-art black-box methods on CUB dataset.

Does not require concept annotations for query samples.

Provides human-interpretable decision-making process.

Abstract

Few-shot learning aims at recognizing new instances from classes with limited samples. This challenging task is usually alleviated by performing meta-learning on similar tasks. However, the resulting models are black-boxes. There has been growing concerns about deploying black-box machine learning models and FSL is not an exception in this regard. In this paper, we propose a method for FSL based on a set of human-interpretable concepts. It constructs a set of metric spaces associated with the concepts and classifies samples of novel classes by aggregating concept-specific decisions. The proposed method does not require concept annotations for query samples. This interpretable method achieved results on a par with six previously state-of-the-art black-box FSL methods on the CUB fine-grained bird classification dataset.