Primacy and recency effects due to momentum in back-propagation

Summary form only given, as follows. Primacy and recency effects are analyzed mathematically for backpropagation algorithms (generalized delta rule), which use momentum. The results show that when the conventional momentum parameter is used, a primary effect occurs: the current values of the weights are biased toward the first presentations in a sequence of training patterns. To produce a recency effect, a different momentum parameter is introduced. The current values of the weights depend more on recent presentations of training patterns under this recency effect. A method is provided for selecting a momentum parameter based on the effect desired: primacy or recency.<>