Systolic Architecture for Higher Order Neural Networks
Jarosław Bilski, Jacek Smoląg · 2003
The higher order neural networks extend classical neural networks structures. In this case we can apply fewer number of layers and neurons to achieve the same results as classical feedforward neural networks. The paper describes the idea of the systolic architecture for higher order neural networks. The systolic architecture for the recall phase is presented and the computational performance is analysed. The number of iteration strongly depends on the network order and number of layer inputs but it does not depend on number of neurons.