Initializing weights to a hidden layer of a multilayer neural network by linear programming

This paper presents a new method for initializing weights to a hidden layer of a backpropagation net. The proposed method initializes weights to hidden units using linear programming such that the backpropagation starts with linear hyperplanes around the input patterns rather than the whole input space. Experimental results show that the method generates the hidden layer much more efficiently and train the net faster than the ordinary backpropagation method.