Relevance score assignment for artificial neural networks

摘要

PROBLEM TO BE SOLVED: To provide information for explaining a reason of a result by an artificial neural network. The back-propagation of the initial relevance score through an artificial neural network redistributes the initial relevance score derived from the network output into a set of items to obtain a relevance score for each item. Backpropagation proceeds to each neuron's set of upstream neighbors according to a broader set of artificial neural networks and distribution functions, so as to compute the preliminary redistributed relevance score of each neuron's set of downstream neighbors. Distributed. [Selection diagram] Fig. 4