Saltus–“A Sudden Transition” Empowered by Federated Learning for Efficient Big Data Handling in Multimedia Sensor Networks

Abstract

In the realm of sensor networks, the substantial rise in multimedia data production, covering audio, video, and acoustic measurements, has expanded the scale of big data. Multimedia Sensor Networks (MSN) excel in managing diverse sensor outputs, representations, and encoding across domains. Existing models for event detection in sensor networks fall short in handling the sheer volume and speed of these measurements from a Big Data perspective. This research work introduces “Saltus,” a model that aligns multimedia data from sensor networks to a standardized feature space. Saltus employs a machine learning-centric architecture to enhance data analysis possibilities. Crucially, the model integrates federated learning to address the evolving landscape of sensor networks. This approach optimizes the collaborative learning capabilities by allowing distributed nodes to train machine learning models locally, preserving data privacy. Saltus emerges as a solution that not only streamlines multimedia data processing but also establishes a more secure and privacy-preserving analytics framework in large-scale sensor networks. The model signifies a step forward in integrating multimedia data into an easily analyzable format, leveraging the advantages of federated learning in big data analytics.

Keywords

• Multimedia sensors
• big data
• machine learning
• federated learning