The Learning of a Back-Propagation Artificial Neural Network

回授類神經網路的學習

翁志祁(Che-Chi Weng)

《華岡工程學報》 16期 (2002/06) Pp. 163-173

https://doi.org/10.29987/CCUHKJE.200206.0009

摘要

本研究建立了一個三層的回授類神經網路，其中輸入層有四個節點，中間層有兩個節點，輸出層有一個節點，以模擬四個輸入的邏輯互斥函數的學習過程。回授類神經網路經過了學習的程序之後，我們觀察到中間層的一個節點X，計算了四個輸入中的任何一對輸入的排列組合是否同時出現爲一的情形，中間層的另一個節點Y，計算了四個輸入中的任何一個是否爲一。輸出層的節點則判斷節點X是否爲僞，且節點Y是否爲真，若兩項判斷同時成立，則此神經網路的輸出爲真，否則爲僞。由此可知，四個輸入的邏輯互斥函數在回授類神經網路中被重新陳述，而爲「若無任何一對輸入的排列組合同時出現爲一的情形，且只要有任何一個輸入爲一，那麼，此邏輯互斥函數神經網路的輸出，則爲真。」

並列摘要

A three-layer back-propagation artificial neural network (BPANN) with four input nodes, two in middle layer nodes, and one output node is constructed for simulating the learning process of performing a four-input logical exclusive-or function. After the training, it is observed that one of the node in the middle layer, say X, calculates any combination of the two inputs are set at the same time. The other node in the middle layer, say Y, calculates any of the input is set. The node in output layer calculates and sets if X is not true and Y is true. The logical exclusive-or of four inputs in the BPANN is reformulated as ”if none of any combination of two inputs are set at the same time and at least one of the input is set, then the output is set”.

並列關鍵字

Back-propagation neural networks ； three-layer neural network ； complete training set ； incomplete training set