D5.3 Continuous Deep Learning Toolkit for real time adaptation

In this document it is described the development carried out in COMPOSITION’s project task Continuous Learning Toolkit for Real Time Adaptation. The document is structured in seven sections and after a comprehensive analysis of the state-of-the-art of relevant fields, a comparison among existing frameworks of interest is then presented. The COMPOSITION project’s use cases in which the task has been involved, are then evaluated and the end users’ historical dataset are assessed through a qualitative based validation process. The main chapter (6) will be the one relative to the analysis of the Deep Learning Toolkit component that has been developed within T5.2. The developed component is going to be deployed in the aforementioned use cases, in which all project’s end users are going to be involved. In this chapter, results will be presented alongside the used methodology. In specific, at first results on extensive tests on synthetic data representing multiple possible scenarios are going to be reported, followed up by real data from one of the end users. At last, in the final section, a comprehensive analysis of the achieved results is then detailed. Furthermore, a detailed report of required work that will be necessary to complete this activity and tackle future challenges is going to be included in the final section.