A Parallel Decomposition Algorithm for
Training Multiclass Kernel-based Vector Machines

Professor Ya-xiang Yuan

Chinese Academy of Sciences
President of Chinese OR Society
J.T. Knight Prize Awardee
L. Fox Prize Awardee
Second Grade Natural Science Prize Awardee
Young Scientist Award of Chinese Academy of Sciences Awardee
Feng Kang Prize of Scientific Computing Awardee
Young Scientist Award of China Awardee
First Grade Prize, Beijing Science and Technology Advancement Award Awardee
Second Grade Prize, National Natural Science Award of China Awardee


Date: 20 January 2009 (Tuesday)

11:30am - 12:30 pm (Preceded by Reception at 11:00 am)

ACC109, Jockey Club Academic Community Centre,
Baptist University Road Campus,
Hong Kong Baptist University


In this talk, I will discuss a decomposition method for training Crammer and Singer's multiclass kernel-based vector machine model. A new working set selection rule is proposed. Global convergence of the algorithm based on this selection rule is established. Projected gradient method is chosen to solve the resulting quadratic subproblem at each iteration. An efficient projection algorithm is designed by exploiting the structure of the constraints. Parallel strategies are given to utilize the storage and computational resources available on multiprocessor system. Numerical experiment on benchmark problems demonstrates that the good classification accuracy and remarkable time saving can be achieved.



The Medium of Instruction: English/Mandarin
All are welcome