The Stereo Neural Network: The Mathematical Architecture and Weight Optimization

This paper proposes a brand new stereo architecture for multilayer perceptron. The proposed stereo neural network (SNN) provides a variety of formation in nodes and the corresponding connections, which can be an architecture to imitate human brain, but with more possibility in the connections of nodes. A mathematical framework is proposed with varying nodes and layers. Also, the method in solving weights in the network is provided. Some offspring structures are examined to exhibit compatible system performance. More possibility is expected to utilize the proposed SNN to seek for better system performance for the future works.