Big Data Dimensionality Reduction at IoT Edge Through Optical Graph Signal Processing

We introduce an innovative edge computing paradigm for big data dimensionality reduction in IoT networks through employing an underlying Graph Signal Processing (GSP) model-alternatively to von Neumann architecture for conventional computing. We realize a new breed of graph filters by instituting photonic wave interference in standard WDM optical communication systems, and thereby, we embody GSP in an optical computation context. Our proposed filter instantaneously takes the average of large-scale multidimensional data while is in transit from IoT edge to its core. Through co-using the communication resources for computation, our work motivates a modern IoT edge computing paradigm based on using a collective pool of computation and communication resources. Our solution saves a significant amount of processing delay, investment in cloud structure, and carbon footprint. It is, therefore, instrumental for breaking”the curse of dimensionality” in complex IoT applications such as self-driving vehicles that involve deploying lots of edge nodes and handling large volumes of high dimensional data with very low latency.