Data processing: artificial intelligence – Machine learning
Reexamination Certificate
2007-06-12
2007-06-12
Hirl, Joseph P (Department: 2129)
Data processing: artificial intelligence
Machine learning
C706S014000, C706S016000, C706S045000
Reexamination Certificate
active
11111031
ABSTRACT:
One embodiment of the present invention provides a system that performs high-level parallelization of large scale quadratic-problem (QP) optimization. During operation, the system receives a training dataset comprised of a number of data vectors. The system first determines to what extent each data vector violates conditions associated with a current support vector machine (SVM). The system then sorts the data vectors based on each data vector's degree of violation. Next, the system partitions the sorted data vectors into a number of prioritized subsets, wherein the subset with the highest priority contains the largest number of violators with the highest degree of violation. The system subsequently solves in parallel a QP optimization problem for each subset based on the subset's priority. The system then constructs a new SVM to replace the current SVM based on the QP optimization solution for each subset.
REFERENCES:
patent: 2006/0112026 (2006-05-01), Graf et al.
R. Collobert and S. Bengio and Y. Bengio, “A Parallel Mixture of SVMs for Very Large Scale Problems”, Advances in Neural Information Processing Systems, 2002, MIT Press url=http://citeseer.ist.psu.edu/collobert02parallel.html.
J. Platt, “Sequential Minimal Optimization: A Fast Algorithm for Training Support Vector Machines” Technical Report 98-14, Microsoft Research, Redmond, Washington, Apr. 1998 url=citeseer.ist.psu.edu/platt98sequential.html.
Pavel Laskov, “Feasible Direction Decomposition Algorithms for Training Support Vector Machines”, Machine Learning, vol. 46, 1-3, publisher=“Kluwer Academic Publishers, Boston”, pp.=“315-349”, year=“2002”, url=citeseer.ist.psu.edu/laskov01feasible.html Only p. 315, 324-325 are furnished with this office action.
Gross Kenny C.
Gurtuna Filiz
Urmanov Aleksey M.
Hirl Joseph P
Park Vaughan & Fleming LLP
Sun Microsystems Inc.
Wong Lut
Yao Shun
LandOfFree
Method for high-level parallelization of large scale QP... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for high-level parallelization of large scale QP..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for high-level parallelization of large scale QP... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3813309