Anti-Noise Relation Network for Few-shot Learning

Few-shot classification has received great attention in the field of machine learning and computer vision. It aims is to achieve the learning ability close to human recognition by training from a few labelled samples. The existing few-shot classification methods have attempted to alleviate the impact of insufficient samples in a variety of ways, such as meta-learning and metric learning, but they ignore the noise robustness. This work proposes a new Anti-Noise Relation Network by embedding an autoencoder network into a classical neural network of fewshot classification, Relation Network. Experimental results on the Stanford Car and CUB-200-2011 datasets demonstrate the superiority of the proposed method in both classification accuracy and robustness against different noises.