Implementing Federated Learning via IoTtalk: A Smart Home Case Study

The Internet of Things (IoT) has enabled many emerging applications through the integration of sensing and communication technologies. Furthermore, with the rapid development of artificial intelligence (AI) in recent years, AI has also been incorporated into IoT systems, giving rise to the concept of Artificial Intelligence of Things (AIoT). AIoT typically includes at least one AI model that assists applications in making intelligent decisions. Users generally need to prepare sufficient data to train these AI models. However, certain IoT applications, such as those in smart homes, have privacy requirements, and users are often unwilling to risk data leakage for the sake of model training. Federated Learning (FL) addresses this issue by offering a distributed learning approach in which clients do not need to upload their raw data. Instead, they upload only the parameters of their local models. The central server then aggregates these parameters to construct a global model, thereby improving overall performance. Nonetheless, implementing a federated learning application requires significant effort from system developers. To reduce this burden and facilitate the rapid development of federated learning applications, we were inspired by the "low-code or no-code (LCNC)" principles promoted by the IoTtalk platform and proposed FLtalk. We implemented a smart home application as a case study to demonstrate the feasibility of FLtalk.