Neural Networks Training Method Using Subnetwork Training algorithms for multi-class prediction neural networks on high level image recognition

摘要

Disclosed are a neural network training method using a partial structure training technique and image data and a device thereof. According to an embodiment of the present invention, the neural network training method comprises the following steps of: operating a neural network output value with respect to each preset multi-dimensional information based on weighed values configuring a neural network with respect to input training data; calculating a training error value based on the neural network output value; dividing the weighed values configuring the neural network by a plurality of clusters; and updating the weighted values, included in the divided clusters, respectively, based on the calculated training error value.