Limited precision incremental communication: error analysis

The effects of the limited precision incremental communication method on the convergence behavior and performance degradation of multilayer perceptrons are investigated. The nonlinear effects of representing the incremental values with reduced (limited) precision on the commonly used error backpropagation training algorithm are analysed. It is shown that the small perturbation in the input(s)/output of a node does not enforce instability. However, when the precision of the incremental values falls below a certain level, the network fails to converge. The analysis is supported by simulation studies of two learning problems