Parity problem, sequential: 1 bit at a time

• Recurrent net learns much faster - even with random weight search: only 1000 trials!

• many fewer parameters

• much better generalization

• the natural solution

Previous slide

Next slide

Back to first slide

View graphic version