Experiments on learning in recursive neural networks

摘要

Summary form only given. An examination is made of training anetwork with recursive connections. The author is able to trainrecursive networks reliably using the generalized delta rule of errorbackpropagation of Rumelhart, Hinton, and Williams on the stationarystates of the recursive network. This method for training recursivenetworks is a truncated form of the recursive error backpropagationalgorithm developed by Pineda and Almeida