Semantic extensions to the Vienna Neural Network Specification Language

The Vienna Neural Network Specification Language (ViNNSL) is an XML based domain specific language for the description of neural network objects. It proved very well as communication framework in service-oriented architecture based neural network simulation environments, as N2Sky. N2Sky is a virtual organization (VO) environment based on the sky computing paradigm. It allows the creation, training, evaluation of artificial neural networks by leveraging arbitrary resources in federated clouds and the exchange of these objects between stakeholders of the VO on a world-wide basis. In this paper we present two novel extensions of ViNNSL to allow a higher degree of knowledge representation. First, we enable ViNNSL to describe the specific application domain of trained neural networks to identify their problem solution capacity. Second, we attribute the description of an instantiated and trained neural network by describing its execution paradigm, sequential and parallel, and the specific necessary infrastructure, hardware and software environment, for processing. These two extensions allow on the one hand for querying and finding neural network instantiations in N2Sky for specific problems, and on the other hand for improving the execution performance of networks by exploiting parallelization approaches transparently to the user.