Advancing Worldwide Navigation Security with Distributed AI
ECE Associate Professor Pau Closas and Assistant Research Professor Tales Imbiriba, in collaboration with Tampere University and the University of Vaasa in Finland, was awarded a $1.4M NSF award for “Distributed AI for enhanced security in satellite-aided wireless navigation (RESILIENT).”
Abstract Source: NSF
The goal of this project is to develop tools for interference management in geolocation applications. The use of Global Navigation Satellite Systems (GNSS) is ubiquitous in civilian, security, and defense applications. As a consequence, the threat of a potential disruption, or even malicious superseding, of GNSS is real and can lead to catastrophic consequences. Therefore, there is a growing need for protecting GNSS against intentional and unintentional interference sources. Particularly, this project investigates a distributed framework to detect and classify threats using hybrid physics- and data-driven models, information which is then used to globally localize the sources of interference. This project is composed of a team of researchers from US and Finland, in a joint effort to advance worldwide security of GNSS against existing threats while providing an excellent opportunity for students to participate in an international project on cutting-edge technologies and methodology development. The team has planned workshops and activities in order to foster a fruitful collaboration between the international research team and students.
This project considers problems related to distributed, collaborative learning tasks, where data-driven AI-models are leveraged to augment physics-based solutions for improved capabilities. The specific goals of the project are divided into three research goals. The first goal investigates the use of deep learning for detection/classification of interference and fusion of multiple correlated classifiers providing local threat detection probabilities. The second goal aims to localize and track the interference sources through the creation of global threat probability maps. This goal is achieved by advancing the field federated learning in a threefold way: (i) to efficiently digesting non-independent and identically distributed data; (ii) combining with active learning methods, whereby moving agents sample specific locations to improve estimation performance; and (iii) investigating federated meta-learning strategies that use task-level knowledge to improve global learning. The third goal of the project investigates the use of those global threat maps to mitigate their effects on collaborative receivers using robust factor graph optimization.