Tackling “leaky” Facial and Object Recognition Operations Over Federated Learning

Federated learning, which decentralizes learning algorithms and data over highly dynamic network and context parameters, presents significant opportunities for computing. Federated learning with facial and object recognition is of particular interest because they are converging on applications with potential privacy risks and vulnerabilities. In this paper, we examine the usage of Internet of Things (IoT) for facial and object recognition; specifically, federated video sensing in monitoring and surveillance. We propose an AI-based approach for federated learning at the edge. Our proposed solution does not transmit any captured data: all personal identifying images and videos are discarded from the IoT device after processing. We suggest that key motifs from the data can instead be framed by AI and traded through the federated learning architecture without added exposure to risk of leak and without saliency loss.