For the BP neural Network algorithm, because it has not been applied to the project, the occasional time of today's study,
The basic idea of this algorithm is this: continuous iterative optimization of the network weights, so that the mapping between the input and output with the desired mapping relationship, using gradient descent method to update the weights of each layer, the objective function to minimize.
1: Initialize the network weights and neuron thresholds (the simplest method is random initialization);
2: Forward algorithm: This is the classical algorithm of BP Neural Network, the main process is to calculate the input and output of the hidden layer neurons and outputs by the formula layer layer.
NET=X1*W1 + x2*w2 + ... +xn*wn
Tanh= (Ex-e-x)/(Ex+e-x)
Input H=tanh (NET);
Hidlayer Compute Prediction Vector P=H*WI
3: Backward algorithm: The calculation results of the forward algorithm, the use of formulas to reverse the correction weights, the output error in some form through the hidden layer to the input layer by layer-to-level inversion, based on gradient descent method to adjust the weight value,
Not to be continued ~ ~
BP Neural Network algorithm