Title :
A multivalent logic approach to risk estimation of learning machines
Author_Institution :
Fac. of Electr. Eng., Maribor Univ., Slovenia
Abstract :
A multivalent logic approach to estimating the risk of error on test samples of learning machine is developed and compared to the bivalent approach based on VC dimension, cover and entropy numbers of sets. The multivalent approach leads to more simple expressions for predicting bounds on the risk estimation which are computable in a short time and use a reasonable amount of computer memory. The results of testing reveal that the multivalent logic algorithm outperforms support vector machines.
Keywords :
learning (artificial intelligence); multivalued logic; risk analysis; support vector machines; VC dimension; Vapnik Chervonenkis dimension; computer memory; learning machines; multivalent logic algorithm; risk estimation; support vector machines; Artificial neural networks; Computer science; Entropy; Informatics; Laboratories; Logic testing; Machine learning; Pattern recognition; Support vector machines; Virtual colonoscopy;
Conference_Titel :
Neural Networks, 2003. Proceedings of the International Joint Conference on
Print_ISBN :
0-7803-7898-9
DOI :
10.1109/IJCNN.2003.1223772
