Abstract
This paper considers the model-based change-point problem with an unknown abrupt change for large-scale network observation sequences. To avoid the difficulty of calculating the normalization coefficients that let the axioms of probability hold, such as in the Exponential Random Graphical Model (ERGM), we present the measure ratio statistics to replace the likelihood ratio statistics. Since the parameter difference reflecting the abrupt change for the network observation sequences is unknown, we first select the dimensions where the parameter difference exists through the -norm penalized maximum likelihood estimation process, then propose a CUSUM multi-chart scheme based on the selected dimensions. Moreover, an optimal design of the CUSUM multi-chart is given when (in-control Average Run length) is large. Two examples are used to illustrate the related theoretical results.
MSC 2010 subject classifications:
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
Note: The first line of each cell is a CUSUM chart with , the second line of each cell is the stochastic CUSUM multi-chart, the last line is the asymptotic optimal CUSUM multi-chart.
Note: The first line of each cell is a CUSUM chart with , the second line of each cell is the stochastic CUSUM multi-chart, the last line is the asymptotic optimal CUSUM multi-chart.
Note: The first line of each cell is a CUSUM with , the second line of each cell is the stochastic CUSUM multi-chart, the last line is the asymptotic optimal CUSUM multi-chart.
Note: The first line of each cell is a CUSUM with , the second line of each cell is the stochastic CUSUM multi-chart, the last line is the asymptotic optimal CUSUM multi-chart.