Transfer Learning on the Edge Networks

Transfer learning focuses on using extensive labeled data samples in the source domain to resolve a different yet related task for the target domain, even when there is no similarity among the training and testing problem’s datasets and distribution of features. This paper will discourse the implementation of the transfer learning model on edge networks to improve the performance factors and communication delay times within different servers. Any extensive system working with embedded systems is considered a high-performance system. An embedded system aims to perform some specific tasks based on the microprocessors, works on low resources and have less power consumption. An embedded system has a functional mapping, and various environment states to generate significant results. For the edge networks, the description of tasks and the dynamics of outer environment is crucial. For further clarification, we developed the transfer learning model. We experimented it on the embedded system using edge device (edge networks) and the local system to compare the time latency of the transfer learning model’s execution. As a result, we concluded that the transfer learning model works effectively and gives us decent accuracy. Implementing a transfer learning model on edge networks is better than implementing on a local system in terms of cost, performance and efficiency.