Authors: Ritika S. Patil, Mohan K. Devre, Afsana Parveen
Abstract: Embedded sensor networks are widely used in smart cities, healthcare monitoring, industrial automation, agriculture and environmental sensing. These networks generate huge amount of data at the edge devices which are usually resource constrained in power, memory and computation. Traditionally, data from sensors is sent to cloud servers for training machine learning models, but this leads to privacy issues, high communication cost, and latency. Federated Learning (FL) is a new distributed learning approach where model training happens locally on devices and only model parameters are shared instead of raw data. This concept is highly suitable for embedded sensor networks where privacy, bandwidth and energy efficiency are critical. This paper reviews the concept of federated learning applied to embedded sensor nodes, discusses system architecture, communication models, challenges, energy constraints, security aspects, and recent applications. Tables and figures are provided to show comparison with traditional centralized learning and architecture of FL-based sensor network. The paper also highlights open research problems and future directions in this area.
Keywords: Federated learning, embedded systems, sensor networks, edge AI, distributed learning, privacy preservation, IoT.
Full Issue
| View or download the full issue | PDF 95-107 |