Bayesian Data Fusion for Distributed Learning скачать в хорошем качестве

Bayesian Data Fusion for Distributed Learning

Rising Star Symposium on Cyber-Physical Systems Security, Resilience, and Privacy https://shorturl.at/C3RZy By Peng Wu from Northeastern University https://jononearth.github.io/ This talk explores two critical pillars of Trustworthy AI—Mixed Reality (MR) Assurance and Federated Learning (FL)—united by a focus on probabilistic methods for safety and privacy. First, we address the safety challenges in mission-critical MR applications, introducing a probabilistic verification framework that leverages Bayesian Networks to model the causal links between system parameters and physiological responses. This approach enables formal safety guarantees and real-time mitigation of cybersickness, moving beyond traditional reactive measures. Second, we tackle privacy and heterogeneity in distributed systems through Bayesian Clustered and Personalized FL, demonstrating how sharing probabilistic posteriors rather than raw data enables effective collaboration in applications such as indoor localization, jammer detection, and multi-agent reinforcement learning. Together, these contributions illustrate a unified path toward AI systems that are rigorously verified for human safety and secure in their data utilization.