A unified format for trained neural network description

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