Resilient Backpropagation

with resilient backpropagation, only the direction of the change of the weights is dependent on current conditions. The magnitude of the change is computed by multiplying this sign by the current . This is the result of multiplying the in the previous iteration by the change in error relative to weight in the last two iterations ( E(t-1)/ w - E(t)/ w). This has the effect of decreasing the delta when the change in error is negative (indicating that a minimum was jumped over) and increasing it if the change is positive (which has an effect of speeding the conversion process).

Resilient Backpropagation

with resilient backpropagation, only the direction of the change of the weights is dependent on current conditions. The magnitude of the change is computed by multiplying this sign by the current . This is the result of multiplying the in the previous iteration by the change in error relative to weight in the last two iterations ( E(t-1)/ w - E(t)/ w). This has the effect of decreasing the delta when the change in error is negative (indicating that a minimum was jumped over) and increasing it if the change is positive (which has an effect of speeding the conversion process).

Back to the Table of Content

Back to the Previous subject

To the Next Subject