Application of Parallel Decomposition for Creation of Reduced Feed-Forward Neural Networks

In this paper a method of creating layers of feed-forward neural network that does not need to be learned is presented. Described approach is based on algorithms used in synthesis of logic circuits. Experimental results presented in the paper prove that this method may significantly decrease the time of learning process, increase generalization ability and decrease a probability of sticking in a local minimum. Further work and goals to achieve are also discussed.