Model Driven Approach for Neural Networks

This paper presents new meta-models for addressing machine learning problems using artificial neural networks. Models conforming to these meta-models can capture the main elements of learning problems and neural networks. This serves as the foundation step for the use of Model-Driven Engineering (MDE) based approach to machine learning using neural networks. The aim is to reap the same benefits which MDE brings to solving software engineering problems. This includes solutions to tool interoperability and standardization challenges, in addition to helping users to develop solutions with less dependence on a particular set of tools and technologies. The presented framework is implemented using Eclipse Modeling Framework (EMF), and several features are demonstrated, including model validation, model transformation, and code generation.