Parallel Java Implementation of a Kernel Ranking Algorithm

B.-J. Falkowski, M. Appelt, C. Finger, S. Koch, and H. van der Linde (Germany)


Machine learning, neural networks, ranking, parallelization.


Large margin perceptron ranking has been proposed in the context of so-called scoring systems used for assessing creditworthiness of banking customers as stipulated in the Basel II central banks capital accord of the G10 states. However, in order to reduce training errors the dual (kernel) algorithm is preferable. Unfortunately it suffers from large CPU time requirements. Hence in this paper a Java parallel implementation is sketched which improves CPU times on the new multi-processor PCs considerably. Encouraging experimental results concerning a Java prototype run on a QuadCore machine are reported.

Important Links:

Go Back