EFFICIENT DATAFLOW MODELING OF MULTILAYER PERCEPTRONS WITH APPLICATIONS TO EVOKED POTENTIAL BASED MEDICAL DIAGNOSIS

摘要

We present an efficient dataflow modeling of Multilayer Perceptron (MLP) algorithm based on the directed graph concept but with controlled global states. More specifically, we investigate the dataflow implementation of two efficient and widely used MLP training algorithms, namely, on-line Backpropagation and Conjugate Gradient in its major versions. The proposed MLP dataflow modeling approach is applied to a medical diagnosis problem, namely, psychiatric case categorization based on evoked potential data classification. The whole system is implement ed in the Labview-G programming environment and it is found that the modeling capabilities along with the results obtained from the proposed MLP dataflow implementation in G demonstrate its efficiency for medical diagnosis applications.