A GPU-Based Training of BP Neural Network for Healthcare Data Analysis

Abstract

As an auxiliary means of disease treatment, healthcare data analysis provides an effective and accurate prediction and diagnosis reference based on machine learning methodology. Currently, the training stage of the learning process cost large computing consumption for healthcare big data, so that the training model is only initialized once before the testing stage. To satisfy the real-time training for big data, this paper proposes a GPU programming technology to speed up the computation of a back propagation (BP) neural network algorithm, which is applied in tumor diagnosis. The attributes of the training breast cell are transmitted to the training model via input neurons. The desired value is obtained through the sigmoid function on the weight values and their corresponding neuron values. The weight values are updated in the BP process using the loss function on the correct output and the desired output. To fasten the training process, this paper adopts a GPU programming method to implement the iterative BP programming in parallel. The proposed GPU-based training of BP neural network is tested on a breast cancer data, which shows a significant enhancement in training speed.