The Upper Bound on the Number of Hidden Neurons in Multi-Valued Multi-Threshold Neural Networks
Nan Jiang, Zhaozhi Zhang, Jian Wang, Xiaomin Ma · 2009
By proposing a computational algorithm, this paper gives the upper bound on the number of hidden neurons to realize multi-valued functions defined on N-points. The architecture of the network is three-layer feedforward neural network with one hidden layer. The network is composed of multi-valued multi-threshold neurons. This upper bound can help us to determine the size of network when we design learning algorithms.