A unified format for trained neural network description

The article introduces a framework for the interchange of trained neural network models. The XML-based language (neural network markup language (NNML)) is presented for the neural network model description, that allows one to write down all components of the neural network model, necessary for its realization. We propose to use the XML notation for full description of neural models, including data dictionary, properties of training sample, pre-processing methods, details of network structure and parameters, method for network output interpretation. The NNML allows interchanging of neural models as well as their documentation, storing and manipulating them independently from the individual simulation system.