-Artificial Neural Network- Chapter 2 Basic Model 朝陽科技大學 資訊管理系 李麗華 教授

Introduction to ANN Basic Model Input layer Hidden layer Output layer Weights Processing Element(PE) Learning Recalling Energy function 朝陽科技大學 李麗華 教授

ANN Components (1/4) Input layer:[X1,X2,…..Xn]t , where t means vector transpose. Hidden layer: I j => net j => Y j Output layer:Yj Three ways of generating output: normalized, competitive output, competitive learning Weights :Wij means the connection value between layers Wij Y ‧ X1 X2 Xn Yj Y1 朝陽科技大學 李麗華 教授

ANN Components (2/4) 5. Processing Element(PE) (A)Summation Function: (supervised) or (unsupervised) (B)Activity Function: or or (C)Transfer Function: Discrete type Linear type Non-linear type 朝陽科技大學 李麗華 教授

ANN Components (3/4) 6. Learning: 7. Recalling: Based on the ANN model used, learning is to adjust weights to accommodate a set of training pattern in the network. 7. Recalling: Based on the ANN model used, recalling is to apply the real data pattern to the trained network so that the outputs are generated and examined. 朝陽科技大學 李麗華 教授

ANN Components (4/4) 8. Energy function: Energy function is a verification function which determines if the network energy has converged to its minimum. Whenever the energy function approaches to zero, the network approaches to its optimum solution. 朝陽科技大學 李麗華 教授

Transfer Functions (1/3) Discrete type transfer function: 1 -1 1 net j > 0 Step function or perceptron fc. Yj= if 0 net j <=0 1 -1 Hopfield-Tank fc. 1 net j > 0 0 net j<0 if Ynj= Yn-1j netj=0 1 -1 -1 net j<=0 1 net j > 0 if Yj = Signum fc. 朝陽科技大學 李麗華 教授

Transfer Functions (2/3) Discrete type transfer function: 1 -1 1 net j > 0 -1 net j<0 if Yj = netj = 0 Signum0 fc. 1 -1 1 net j > 0 -1 net j<0 if Ynj = net j = 0 Yn-1j BAM fc. 朝陽科技大學 李麗華 教授

Transfer Functions (3/3) Linear type: Nonlinear type transfer function: Yj = net j net j net j > 0 Yj = if 0 net j <=0 Yj = Sigmoid function Yj = Hyperbolic Tangent function 朝陽科技大學 李麗華 教授

Energy function (a) The energy function for supervised network learning: E = where E is the energy value ΔW= this is the value for adjusting weight Wij (b) The energy function for unsupervised network learning: E = ΔW= this is the value for adjusting weight Wij 朝陽科技大學 李麗華 教授