Research On Computation Methods Of Neural Networks And Applications Based On A Cloud Computing Platform

With Internet gaining its popularity, large data is generated every day. Cloud Computing provide a solution with high availability, high throughout and high cost performance ratio for the effective usage of these data. Due to Neural Network’s built-in propriety of parallelism, it’s implementation is friendly to cloud computing, which attract many researcher working on it. However, most of these researches focus on Back Propagation Neural Network, research on other kinds of neural network like Radial Basis Function Neural Network is rarely, and the paralleled neural network has not been used in any specified application.With RBF Network’s (shorten for Radial Basis Function Neural Network) ability of approximating any nonlinear function, it is a good solution to deal with system which regularity is difficult to resolve. RBF Network has strong generalization ability and a fast convergence rate for the learning, it has been successfully applied to many areas such as nonlinear function approximation, pattern recognition and image processing. In the tradition RBF Network’s algorithm, all the training examples will be loaded to the network, which make the algorithm difficult to scale. In this paper, we solve the problem by applying RBF Network to Cloud Computing. The algorithm’s practice can be concluded like this:1. introduce HDFS to store the training examples;2. every datanode handle the examples stored on it;3. the namenode collect adjustment from datanodes and applied the adjustment to the network;4. Run1-3multi times until convergence is matched.Face Recognition is the technology that applies analyzing and comparing human face features to authentication, and it’s widely used in many areas. After many years research, this kind of technology has been fully developed. Thus face recognition algorithm is implemented with neural network based on cloud computing, and comparison is made between this paralleled algorithm and the serial algorithm, which purpose is to verify the effectiveness and advantage of neural network based on cloud computing platfrom.Voice Affection Recognition is the technology that analyze voice signal to recognize speaker’s affection. Tradition voice signal analyze technology focus on speech content recognition and speaker recognition, but research of voice affection recognition, especially voice with Chinese background, is a new research area. In Voice Affection Recognition, massive calculation is needed for the pretreatment of voice signal, which can be paralleled with cloud computing platform. In this paper, we combine voice signal processing and cloud computing platform, implement a voice affection recognition algorithm with neural network based on cloud computing platform. For smart computers and robots, just having the ability to recognize emotion is not enough to provide a harmonious human-computer interaction environment, they need to be able to respond to human emotional states, in another way, they need to have the emotional capacity. To ahieve Affective Computing, the establishment of emotional intelligence model is required. With references of previous work, this paper give a three-layerd emotional model based on hidden Markov model, meanwhile a simulation model combined the cloud computing platform and the emotion model.