Run-off election-based decision method for the training and inference process in an artificial neural network.

摘要: Generally, the decision rule for classifying unstructured data in an artificial neural network system depends on sequence results of activation function determined by vector-matrix multiplication between input bias signal and analog synaptic weight quantity each node a matrix array. Although sequence-based can efficiently extract common feature large set short time, it occasionally fail to classify similar species because does not intrinsically consider other quantitative configurations that affect update. In this work, we implemented simple run-off election-based via additional filter evaluation mitigate confusion from proximity output functions, enabling improved training inference performance system. Using selected difference among features classified images, recognition accuracy achieved three types shoe image sets reached ~ 82.03%, outperforming maximum ~ 79.23% obtained fully connected single layer network. This algorithm with independent precisely supply class step