Few works tackle this problem by exploiting sample correlations. In this paper, we propose BundleNet, a framework of sequential structure (named bundle-module, see Fig. 1) for deep neural networks to ...