The contents of this article for I learn to understand, there is wrong place also please point out.
The so-called BP neural Network (back propagation) is to use the known data set along the neural network forward to calculate the predicted value, so as to obtain the deviation between the predicted value and the actual value, and then use this deviation and the neural network deviation gradient descent direction to adjust the weight parameter between the layer and the layer, so as to reduce the deviation. So constantly forward to calculate the deviation, and then along the direction of deviation to adjust the weight parameters backward, until the deviation value down to a permissible degree, you can train to obtain a BP neural network.
In this sense, the BP neural network is dependent to a certain extent on the data set. The better the quality of the data set, the more the number of data sets, the better predictive of the trained BP neural network.
Start with the simplest two-tier network below
The network can actually be seen as a linear fit for input layer data.
BP Neural network