Speed up method for neural network learning by using GPGPU

Neural network is a mathematical models for machine learning methods. This model apply to the many types of classification problem. And recently, many applications with neural network are required to process a big data in real time. In this paper, we discuss how to make the processing time of the neural network learning faster by using GPU. GPGPU is a technique by which GPUs are used for a general computation approach. GPU is a dedicated circuit to draw the graphics, so it has a characteristic in which many simple arithmetic circuits are implemented. This characteristic is applied to not only graphic processing but also general purpose like this proposed method. In order to employ it effectively, the calculation of the neural network learning are implemented to process simultaneously. The calculations which the neurons in the layer and many patterns are processed is parallelized. And we propose the parallelize method for calculation of back propagation. As the result, the proposed method is 25 times faster than the non-parallelized.