Parametric Optimization of Reconfigurable Designs Using Machine Learning
This paper presents a novel technique that uses meta- heuristics and machine learning to automate the optimization of design parameters for reconfigurable designs. Traditionally, such an optimization involves manual application analysis as well as model and parameter space exploration tool creation. We develop a Machine Learning Optimizer (MLO) to automate this process. From a number of benchmark executions, we automatically derive the characteristics of the parameter space and create a surrogate fitness function through regression and classification. Based on this surrogate model, design parameters are optimized with meta-heuristics. We evaluate our approach using two case studies, showing that the number of benchmark evaluations can be reduced by up to 85% compared to previously performed manual optimization.
Keywordsoptimization surrogate modeling PSO GP SVM FPGA
Unable to display preview. Download preview PDF.
- 3.Su, G.: Gaussian process assisted differential evolution algorithm for computationally expensive optimization problems. In: PACIIA, pp. 272–276. IEEE Computer Society (2008)Google Scholar
- 4.Guoshao, S., Quan, J.: A cooperative optimization algorithm based on gaussian process and particle swarm optimization for optimizing expensive problems. In: CSO, vol. 2, pp. 929–933 (2009)Google Scholar
- 6.Kurek, M., Luk, W.: Parametric Reconfigurable Designs with Machine Learning Optimizer. In: FPT (2012)Google Scholar
- 9.Rasmussen, C., Williams, C.: Gaussian Processes for Machine Learning. MIT Press (2006)Google Scholar
- 10.Bishop, C.M.: Pattern Recognition and Machine Learning. Springer (2006)Google Scholar
- 11.Van Den Bergh, F.: An analysis of particle swarm optimizers. Ph.D. dissertation, University of Pretoria, South Africa (2002)Google Scholar