Deep InterBoost networks for small-sample image classification

Deep neural networks have recently shown excellent performance on numerous image classification tasks. These often need to estimate a large number of parameters and require much training data. When the amount data is small, however, network with high flexibility quickly overfits data, resulting in model variance poor generalization. To address this problem, we propose new, simple yet effective ensemble method called InterBoost for small-sample classification. In phase, first randomly generates two sets complementary weights which are used separately base same structure, then updated refining through interaction between previously trained. This interactive process continues iteratively until stop criterion met. testing outputs combined obtain one final score Experimental results four datasets, UIUC-Sports, LabelMe, 15Scenes Caltech101, demonstrate that proposed outperforms existing ones. Moreover, from Wilcoxon signed-rank tests show our statistically significantly better than methods compared. Detailed analysis also provided an in-depth understanding method.