Distributed Edge Intelligence
Artificial Intelligence (AI) is one of the most influential milestones in the History of Humanity. However, AI engines need many data, including those from IoT devices, to become even more powerful. Indeed, in the next-Internet generation, humans are integrated into it (Human-in-the-Loop - HiTL -concept), such as in intelligent clothes, personal smartphones, smart watches, etc. Thus, the current AI models need more data to estimate behavior. However, many data need higher execution times and energy consumption, both unavailable to edge devices. Due to this, the current models use a central server in the cloud and select the most potent edge devices, e.g., in Federated Learning computation.
Federated Learning (FL) is a distributed Machine Learning (ML) for decision-making that hides sensitive data from different users. Each federated client computes a model update based on global variables. The central server maintains updates without exposing raw data among participants.
The central question is related to the need for data training. What data size is needed? Could IoT devices with appropriate hardware be used to provide edge intelligence? Our research commits to identifying new AI algorithms and methods for overcoming these issues.