For hidden layers of two sub-cases discussed, and the conversion point by the formula y = Sigma (wx) represented (canceled constant term + b):
No hidden layer: Assuming that the input layer has two nodes, an output layer nodes. At this point there are two weights and wires w1 w2, these two parameters determine a linear analytic, and therefore geometric meaning of the neural network is to represent a straight line.
A hidden layer: Suppose there are two input nodes, 2 hidden layer nodes, an output layer nodes. In this case there are four parameter weights from the input layer to the hidden layer, which may be a 2 * 2 matrix to represent the geometrical meaning of this is to represent a transformation matrix. It may be identical transformation, translation can also be transformed and so on. E.g. w1, w2, w3, w4 is 1,0,1,0, respectively, representing an identity transformation. At this time, the input layer nodes 2 parameter represents a point. Geometric meaning of this neural network is a point through a geometric change position after conversion into a new point (hidden layer), and finally there is a straight line separated.