![]() | Only 14 pages are availabe for public view |
Abstract This thesis is concerned with the design of two types of artificial neuron models based on a polynomial architecture. The first neuron model is called the generalized mean single multiplicative neuron (GMMN) and the second neuron model is called the geometric mean single multiplicative neuron (GEOMN).The architecture of the two neuron models has been presented with their entire connections and learning algorithm in chapter 3 and 4. The mathematical model in representation of SMN model and its learning algorithm was introduced in chapter 2 that reflects the characteristics of it. We used a standard backpropagation (BP) algorithm [32] to train the two proposed neuron models, which is based on the steepest descent gradient. The GMMN, GEOMN and SMN models are implemented in MATLAB. The computational power and approximation capability were demonstrated through a solid set of simulations and performance evaluation metrics. |