Infer real structure of neural network by using method of partial causality
Chen Song-qiao · Computer Engineering and Applications Journal · 2007
Since Granger have put forward the notation of causality,the application of Granger causality in the construction of the structure of neural network has become more and more extensive,we can get a graph with direction from them.For a network with two neurons,can analyze who is the cause and who is the result by using ordinary Granger causality.For a network with more than three neurons,can’t analyze it with the ordinary Granger causality,for there exists indirect effect between them.We should analyze it with partial causality.This paper introduces the basic notation of partial causality,and also compare the difference between ordinary Granger causality and partial causality in a simulative network with three neurons.