A new learning algorithm, the progressive learning algorithm, is propo
sed for use in a goal-seeking feedforward architecture, GSN(f). Experi
ments show that both improved recognition performance and significantl
y faster training times can be achieved with this learning algorithm.