Optimal solution of linear inequalities with applications to pattern recognition
An algorithm is developed for the optimal solution of systems of linear inequalities, where the criterion is the number of inequalities in the system which are satisfied. In the terminology of pattern recognition, the algorithm computes a linear decision function which minimizes the number of misclassified patterns in the training set. The method is a search procedure modeled, up to a point, on a previous algorithm of Warmack and Gonzalez. Some new ideas are introduced, however, which greatly improve the efficiency of the new algorithm. Experiments are performed which demonstrate the efficiency of the algorithm, and comparisons are made with the algorithm of Warmack and Gonzalez and with other algorithms. Finally, a lower bound is derived for the number of operations of a certain type which must be carried out.
Thesis80.C437.pdf
2.94 MB
Adobe PDF
2d8f2b095cf634ee47a16de6e0f5d3dd