A Review on Edge Intelligence based Collaborative Learning Approaches

Abstract

Edge Intelligence (EI) based collaborative learning approaches have been researched in the recent years by several researchers. As majority of existing collaborative learning approaches are designed in the context of servers, design and development of novel systems are required in order to apply collaborative learning in resource constrained devices. Researchers possess different interpretations of the existing collaborative learning approaches. This paper presents an analysis of the existing work conducted on collaborative learning approaches on the domain of EI. The strongholds and drawbacks of the existing work will be analyzed to identify an ideal collaborative learning approach to be applied on the Internet of things (IoT) edge. The analysis will further include an investigation into the future enhancements and research gaps. The partitioned model training approach has been identified as the most ideal approach for the IoT edge based on the conducted critical analysis process. The reduction of communication overhead in the partitioned model training approach and the application of the partitioned model training approach in other unexplored deep learning model architectures have been identified as future research directions. This paper is a work in progress section of an ongoing research and in the future the findings will be used to design a system to apply the collaborative learning approach in IoT Edge and bridge the existing research gaps.