Support Vector Machine 2002 2 1 SVM n

  • Slides: 33
Download presentation
Support Vector Machine 支持向量机 张鑫 2002. 2. 1

Support Vector Machine 支持向量机 张鑫 2002. 2. 1

SVM的分解算法 n Edgar Osuna(Cambridge , MA)等人在 IEEE NNSP’ 97发表了An Improved Training Algorithm for Support

SVM的分解算法 n Edgar Osuna(Cambridge , MA)等人在 IEEE NNSP’ 97发表了An Improved Training Algorithm for Support Vector Machines , 提出了SVM的分解算 法

SVM的分解算法 n Proposition(Build down): moving a variable from B to N leaves the cost

SVM的分解算法 n Proposition(Build down): moving a variable from B to N leaves the cost function unchanged, and the solution is feasible in the subproblem n Proposition(Build up) moving a variable that violates the optimality condition from N to B gives a strict improvement in the cost function when the subproblem is re-optimized

SVM的分解算法 1. 2. 3. Arbitrarily choose |B| points from the data set. Solve the

SVM的分解算法 1. 2. 3. Arbitrarily choose |B| points from the data set. Solve the subproblem defined by the variable in B. While there exist some j j N, such that replace any i , i B, with j and solve the new subproblem.

SVM分解算法的实例 n SVMlight Thorsten Joachims (University. Dortmund , Informatik, AI-Unit) Make Large-Scale SVM Learning

SVM分解算法的实例 n SVMlight Thorsten Joachims (University. Dortmund , Informatik, AI-Unit) Make Large-Scale SVM Learning Practical n SMO John C. Platt (Microsoft Research) Fast Training of Support Vector Machines using Sequential Minimal Optimization