SVM Regression Model For Design Space Exploration Of Micro-architecture

In recent years, with the development of semiconductor technology, the processor chips have highly integrated, processing speed of single chip have approch the limit.The technology of Multi-core has become a hot research topic for improve the performance of processors, but how to collaborative work with other chips, The multi-/many- core processors will result in sophisticated large-scale architecture substrates that exhibit increasingly complex and heterogeneous behavior., This also brings to the architecture design a new research project, Existing methods lack the ability to accurately and informatively forecast the complex behavior of micro-architecture across the design space such as cycle simulation and traditional method., In the beginning of the micro-architecture design, to predict their behavior is particularly very important.Support vector machine is a learning machine method which is the most widely used recently, based on statistical learning theory. Suitable for small sample study, has a strong generalization performance, the other prediction models will be difficult to resolve the problems of local minima and the curse of dimensionality. But how to choose the parameters of model is still not a standard of theoretical knowledge as a guide, in the past, designers have mostly rely on their own prior knowledge for the design. Therefore, to achieve good design results requires a lot of experience as a guarantee, it is difficult to obtain good results, and greatly affects the support vector machine method of a wide range of promotion, The genetic algorithm is applied to build support vector machine regression model parameters of the adaptive optimization algorithm, using genetic algorithm to optimize the support vector machine model of the important parameters of design.Support vector machine has been successfully applied in many fields, but most are classification problems, so researches in regression estimation based on SVM need to be enhanced. The important point of this study is that find a regression prediction model which based on support vector machine and better than the neural network method, In the beginning of the micro-architecture design to predict its workload. By the genetic algorithm, support vector machines'parameters could be optimized, we could receive the support vector machine regression prediction model, and experiment based on LIBSVM, compared the forecast results of neural network with support vector machine prediction results, it is evident that support vector machine regression model instead of RBF neural network structure in the wavelet parameter prediction can achieve better results. Therefore, the contents of this thesis can be helpful to relative researches or applications.