Mixture Gaussian Prototypes for Few-Shot Learning
Ruijin Jiang, Zhaohui Cheng · 2021 International Conference on Data Mining Workshops (ICDMW) · 2021
In this paper, we provide a new model named Mixture Gaussian Prototypes in few-shot classification problems. In order to describe their features more accurately, our mixture Gaussian prototypes use both the variance and the mean of the data in the support set, while the Gaussian distribution is concise so that it is suitable for the small sample task. Moreover, we verify the validity of our mixture Gaussian prototype through experiments. Also, we perform experiments on Omniglot, mini-ImageNet, and CUB for few-shot Classification and obtain high classification accuracy.