• 作者： 黎自奮; 張智立
• 作者服務機構： 國立中興大學應用數學系; 樹德工商專科學校數學科
• 中文摘要： 在分類上，我們假定每一類的條件密度函數是常態分配，但其參數是未知，而且每類的百分比(prior probability)也是未知。在本論文中，我們將提出一個遞迴式子(recursive formula)的演算法，利用一組未分類的圖形，去求得所有這些未知參數的估計值，而這些估計值都將會收斂到其參數值。此外，我們用電腦模擬後，發現此方法能得到不錯的結果。
• 英文摘要： In classification, we assume that the conditional density function for each class has a normaldistribution, and we consider the most general case where all the parameters in all classes are assumedto be unknown; i.e., the prior probability of each class and all the parameters in the conditional normaldensity functions are unknown. An algorithm for a recursive formula is presented to estimate all theunknown parameters using a set of unidentified input patterns. The estimates obtained by the algorithmall converge to their true unknown parameters. The results of a Monte Carlo simulation study with normaldistributions are presented to demonstrate the accurate estimation of unknown parameters in all classes·
• 中文關鍵字： Bayes rule; classification; pattern recognition; unsupervised learning
• 英文關鍵字： --