DETERMINING INFLUENCE OF ATTRIBUTES IN RECURRENT NEURAL NET-WORKS TRAINED ON THERAPY PREDICTION

摘要

A method and system of determining influence of attributes in Recurrent Neural Networks (RNN) trained on therapy prediction is provided. For each output neuron z_k^l a relevance score R_k^l is decomposed into decomposed relevance scores R_k→j^l for each component x_j^l of an input vector x¹ and all decomposed relevance scores R_k→j^l of the present step l are combined to a relevance score R_j^l for the next step l−1.