作者: Włodzisław Duch , Rafał Adamczak , Krzysztof Grąbczewski
关键词: Function (mathematics) 、 Computer science 、 Feature selection 、 Backpropagation 、 Complex system 、 Computational intelligence 、 Forcing (recursion theory) 、 Artificial neural network 、 Logical rules 、 Artificial intelligence 、 Data mining
摘要: Three neural-based methods for extraction of logical rules from data are presented. These facilitate conversion graded response neural networks into performing functions. MLP2LN method tries to convert a standard MLP network operations (LN). C-MLP2LN is constructive algorithm creating such networks. Logical interpretation assured by adding constraints the cost function, forcing weights ±1 or 0. Skeletal emerge ensuring that minimal number found. In both covering many training examples generated before more specific exceptions. The third method, FSM2LN, based on probability density estimation. Several performance these