Application-specific transfer learning over edge networks

Abstract

Transfer learning uses a profound labeled set of data from the source domain to deal with a similar problem for the target domain. Transfer learning provides accurate decision- making when insufficient data samples are available and when building a new prediction model takes more time and effort. This study explains comparative analysis of traditional machine learning techniques and transfer learning approaches over edge networks to enhance the performance and networking latency within discrete nodes. Edge networks are widely used to improve the efficiency and staging of any algorithm as the embedded systems focus on implementing some particular events based on the microprocessors and, at the same time, working on the least resources that result in having less power consumption. Moreover, we generated a hybrid-based transfer learning model to avoid negative transfer. This thesis uses two case studies: mushroom sales prediction and heart attack detection system.