學習案例集規則之推導--改良式 L1迴歸方法的運用

Abstract

本文運用改良式 L1迴歸法來推導大型學習案例中的規則 .案例式學 習法可分為 單階學習法與多階學習法 ,單階式案例學習法 (AQ 法)與多 階式案例學習法 (類神經網路法 )的缺點為不易求得最佳解及不易加入限 制條件 . 本文的目的及在發展出一套新的案例學習法 ,以在合理的計算 時間下求得最佳解 .文中並以動物園裡動物分類為例,進行案例學習規則 的推導 , 並證實多階學習法可較單階學習法求得更佳解. This thesis uses a modified L1 regression method to induce rules from large learning examples. The deficiencies of current learning methods (AQ and Neural Networks ) are : firstly , they can only find local optimum ; secondly , they are not allowed to add constraints. This thesis proposes a new method to deduce optimal rules in reasonable computation time. Ananimal classification instance verifies that the multiple layers learning method is better than the single layer learning method.