- 作者: 李宗寶
- 作者服務機構: 國立中興大學應用數學系
- 中文摘要: 對一序列獨立的指數群(exponential family)隨機變數,我們將偵測改變點的個數以及改進此偵測的結果。改變點閒的限制如 (方程式無法摘錄) 能被去除和罰則項(penalty term)如d(n)能被改進到log n。此篇也推廣二項式分佈的結果。其主要的重點在證明任何對數概似比(log likelihood ratio)的部分和之最大值除以log n將機率收斂(convergence inprobability)至1。
- 英文摘要: The nurnose of this work was to improve the detection for the number of chance points in a sequenceof inaepenaent exponentiat iamiiy ranaom variames· The restriction (方程式無法摘錄) between the cnange pointscan be dropped, and the penalty term d(n) can be improved to be logn.The result can extend the binomialcase.The key point for the improvement is that the maximum of any partial sum of the log likelihoodratios divided by logn will be shown to converge to one in probability.Key Words:change points, Schwarz’s criterion
- 中文關鍵字: change points; Schwarz's criterion
- 英文關鍵字: --