Supervised learning on a fuzzy Petri net.

Feed-forward neural networks used for pattern classification generally have one input layer, one output layer and several hidden layers. The hidden layers in these networks add extra non-linearity for realization of precise functional mapping between the input and the output layers, but semantic relations of the hidden layers with their predecessor and successor layers cannot be justified. This paper presents a novel scheme for supervised learning on a fuzzy Petri net that provides semantic justification of the hidden layers, and is capable of approximate reasoning and learning from noisy training instances. An algorithm for training a feed-forward fuzzy Petri net and an analysis of its convergence have been presented in the paper. The paper also examines the scope of the learning algorithm in object recognition from 2D geometric views. [ABSTRACT FROM AUTHOR]