Speeding up Discriminative Learning Quadratic Discriminant Function with Sampling

摘要

The Discriminative Learning Quadratic Discriminant Function (DLQDF) is a favorable method for multi-classification problems, especially for handwritten character recognition, due to its prominent classification effects. However, its training complexity is very high. To improve the efficiency of DLQDF while keeping its supper performance, this paper presents a sampling method - MBS (MQDF-Based Sampling) to speed up the process of parameter learning. Experiments show that MBS can effectively speed up the training of DLQDF while keeping classification accuracy.