Counting Cycles in Networks

Jiashun Jin (Carnegie Mellon University)

For many years, researchers have been looking for a testing statistic that is powerful and also has a tractable limiting null. We show that such a statistic can be constructed using the network cycle counts. We show that the statistic can be useful for network goodness-of-fit and for estimating the number of network communities. We also discuss how to properly address the minimax lower bound in network testing, using Sinkhorn's matrix scaling theorem. For applications, we show the statistic can be useful in measuring research diversity and in building co-authorship tree.


Back to Day 1