DSL approach to deep learning lifecycle data management

A new approach to Deep Learning (DL) lifecycle data management tool support is presented: a very simple DL lifecycle data management tool, which however is usable in practice (it will be called Core tool) and a very advanced extension mechanism for this Core tool which in fact converts the Core tool into a DSL tool building framework for DL lifecycle data management tasks. The extension mechanism is based on the metamodel specialisation approach to Domain Specific Language (DSL) modelling tools introduced by the authors. The main idea of metamodel specialisation is that we first define the Universal Metamodel (UMM) for a domain and then for each use case in the domain define a Specialised Metamodel (SMM). The paper concludes with a detailed description of future research directions, concerned with defining a more general UMM and its usage.