Generalized Adaptation for Few-Shot Learning

TL;DR

This paper introduces a closed-form base learner for few-shot learning that enhances generalization, supported by theoretical analysis and state-of-the-art experimental results across multiple benchmarks.

Contribution

It proposes a novel closed-form base learner that constrains adaptation, improving generalization in few-shot learning, with theoretical validation and superior experimental performance.

Findings

Achieves 87.75% accuracy on 5-shot miniImageNet

Outperforms existing methods by approximately 10%

Validates the approach with theoretical analysis

Abstract

Many Few-Shot Learning research works have two stages: pre-training base model and adapting to novel model. In this paper, we propose to use closed-form base learner, which constrains the adapting stage with pre-trained base model to get better generalized novel model. Following theoretical analysis proves its rationality as well as indication of how to train a well-generalized base model. We then conduct experiments on four benchmarks and achieve state-of-the-art performance in all cases. Notably, we achieve the accuracy of 87.75% on 5-shot miniImageNet which approximately outperforms existing methods by 10%.