Training

Neural network's Training may be distinguished by weather training patterns are represented incrementally or in batches. In incremental training, weights are updated after the representation of every training input-output pairs. An error measure, whole pattern sum of squares error or PSS, is determined for iteration as:

                                                    .................Eqn(24)

Whereas

-          o is the number of output units,

-          And T_j is the training values for the output node

-          And activation, A_j is respectively, for the output node.

The training pattern is repeatedly represented until a tolerable error level is obtained, determined by as:

                                                            E ≥ TSS..........................Eqn(25)

Whereas E is the some usual criterion

In batch training, a set of pattern is represented to the network. The network error terms for every pattern are summed, and merely at the end of presentation of all the patterns are the weights updated. TSS, falls beneath E, as specified by as:

                                                       ................Eqn(26)

Whereas p is the number of patterns in the batch

Whenever the incremental approach much closely mimics the learning experience, this tends to perform more poorly than the batch approach. The system quality response becomes dependent on the order whether illustrations are presented, and the extent also to which error is minimized for each pattern. Minimization of error for an exact pattern may cause increased response error for other pattern. The batch approach minimizes response error over all illustration patterns, resulting in better overall performance. Though, a complete set of illustrations covering all contingencies will rarely be available for a realistic application. Some combination therefore of batch and incremental modes might be helpful.