SimpleShot: Revisiting Nearest-Neighbor Classification for Few-Shot Learning

TL;DR

This paper demonstrates that simple feature transformations combined with nearest-neighbor classification can achieve competitive few-shot learning results, challenging the reliance on complex meta-learning methods.

Contribution

The study shows that basic feature transformations like mean-subtraction and L2-normalization significantly improve nearest-neighbor classifiers for few-shot learning without meta-learning.

Findings

Nearest-neighbor with feature transformations outperforms prior methods in 3 of 5 miniImageNet settings.

Simple transformations can match or surpass complex meta-learning approaches.

Neural network features with minimal processing are effective for few-shot classification.

Abstract

Few-shot learners aim to recognize new object classes based on a small number of labeled training examples. To prevent overfitting, state-of-the-art few-shot learners use meta-learning on convolutional-network features and perform classification using a nearest-neighbor classifier. This paper studies the accuracy of nearest-neighbor baselines without meta-learning. Surprisingly, we find simple feature transformations suffice to obtain competitive few-shot learning accuracies. For example, we find that a nearest-neighbor classifier used in combination with mean-subtraction and L2-normalization outperforms prior results in three out of five settings on the miniImageNet dataset.