The transverse network-a new neural model with spatio-temporal coding of information

It is the purpose of this paper to provide a new neural network architecture based upon an efficient spatio-temporal coding scheme. The way the data are processed by the neural network allows the classification of either static or dynamic information. The neural network the authors propose includes local short term memory at the cell level and a long term memory at the cell´s group level. The learning strategy is a supervised one, it is based on a temporal, local and real-time version of the backpropagation algorithm. The constraint of a layered architecture imposed by the original algorithm is suppressed, making it possible to use recurrent networks.