Federated Learning is gaining more prominence and has become increasingly popular in the present age, primarily impacted by the pandemic, where dependence on devices has increased tremendously due to social distancing, lockdown measures, limited human mobility, and accessibility. Its impact on the healthcare industry has been the largest, given the fact that the healthcare domain has a lot of sensitive and private data of individuals. Further with deployment and use of IoT sensor devices becoming easier, Federated Learning (FL) based systems have contributed much in human health monitoring and discovering new trends, patterns, and anomalies from the human body. The IoT sensor devices used in FL architectures are intelligent and time-sensitive that can send notifications to users based on any change in user’s health conditions, or sudden changes in the environments, that might unfavorably impact the user’s health.
This talk first introduces the audience to few use-cases in the healthcare industry where Federated Learning-based systems can be used. In the next phase, our talk demonstrates a three-layer network-edge driven federated learning-based generic learning system framework that extracts client features from the devices engaged in collecting the data before training and sharing their locally trained models to the edge and central global server. The system is further endowed with a real-time monitoring pipeline that incorporates model governance and ensures fair predicted outcomes to end-users without exposing their private confidential information. In addition, it provides a detailed overview of different KPI metrics that can be used to test the robustness of an ML, the fairness aspect of it on different minority sub-groups. In this context, the talk exemplifies the merits of selecting the right model KPIs, from the standpoint of Ethical and Sustainable AI, when the model is trained using data from several heterogeneous IoT devices and deployed in a scalable enterprise-grade deployment.