Regular Grammar Inference - A Hybrid Connectionist/Symbolic Approach

Short Description

Dynamic Recurrent Neural Networks have been successfully used to learn regular grammars. An efficient algorithm for training second order RNN will is presented. In addition, recent advances have made it possible to extract finite state automata from the trained neural network. An approach based on hierarchical clustering will be explained.

Reference

A Hybrid Connectionist Symbolic Approach To Regular Grammar Inference Based on Neural Learning and Hierarchical Clustering.
R. Alquezar and A. Sanfeliu.
In the proceedings of the Second ICGI'94, Spain. pp. 203-211

Experimental Comparison of the Effect of Order in Recurrent Neural Networks.
Clifford Miller and C. Lee Giles
In the International Journal of Pattern Recognition and Artificial Intelligence Vol. 7. No. 4 (1993) pp 849-872