Support Vector Machine Parameter tuning using Dynamic Encoding algorithm for handwritten digit recognition

In this paper, we propose a support vector machine parameter tuning algorithm using dynamic encoding algorithm for handwritten digit recognition. This method uses dynamic encoding algorithm for search (DEAS) which is recently proposed optimization algorithm based on variable binary encoding length. The radius/margin bound is used for the estimation of the support vector machine generalization performance. When the radius/margin bound is not convex form or different from real error rate for test data set, n-poled validation error rate can be used for parameter tuning. The proposed method can be applied to the case which is hard to find gradient information of radius/margin bound. Moreover, the proposed method is an more efficient algorithm compared with GA algorithm and grid search in computation time.