1970/01/01-1970/01/01
Abstract
We are inspired by an application in deep learning called the DropConnect method, which sparses edges in a complete bipartite graph (connections) to avoid overfitting, and its application to experimental design to improve estimation accuracy. We propose a combinatorial problem called Spanning Bipartite Block Design (SBBD) and show how to construct designs that satisfy the combinatorial requirements of SBBD.
数学系主页
© 2015 All Rights Reserved. 粤ICP备14051456号
南方科技大学数学系