MPPCANet: A feedforward learning strategy for few-shot image classification

Highlights：

• The basic assumption that the image patches are sampled from a single Gaussian distribution is eliminated. A multiple Gaussian distributions assumption is adopted.

• The mixtures of the probabilistic principal component analysis (PPCA) model is used to model the image patches. The parameters in the model is estimated using the expectation-maximization (EM) algorithm. The clustering of the image patches and the principal components of each cluster is simultaneously obtained during the parameter estimation procedure.

• A detailed theoretical comparison between the CPCANet proposed in our previous work and the MPPCANet is given. The theoretical analyzes are given in the three subsections.

• The proposed method improves the generalization ability of existing feedforward learning networks under the few-shot experiment setting.