Export Reviews, Discussions, Author Feedback and Meta-Reviews

"NIPS Neural Information Processing Systems 8-11th December 2014, Montreal, Canada",,, "Paper ID:","1915" "Title:","Learning on graphs using Orthonormal Representation is Statistically Consistent" Current Reviews First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. The authors consider the problem of node classification in graphs, in the transductive setting. For the node classification problem, they argue that orthonormal representation of graphs, i.e. embedding of graphs onto a sphere, where two nodes with no edge between them correspond to two orthonormal vectors on the sphere, is a consistent representation, when used with a certain regularized formulation that is popularly used for vertex labeling. The proof proceeds by considering the function class, learned by the regularized formulation, and using fairly standard techniques to establish generalization bounds. Bounding the Rademacher complexity of this function class, allows one to provide generalization error bounds.