A digital art illustration of a cityscape with circuit lines, a control panel, and glowing icons representing cybersecurity and data on a screen, including a padlock symbol.

This CAREER project addresses a critical challenge in distributed learning systems: developing secure, robust, and privacy-preserving machine learning systems for infrastructure essential to society. As critical infrastructures, such as power grids, increasingly rely on data-driven decision-making, machine learning plays a central role in optimizing operations. However, deploying machine learning in these settings introduces severe privacy and security risks. Recent cyberattacks, including data poisoning and model inversion, have exposed vulnerabilities that could lead to widespread disruptions. Federated Learning (FL) has emerged as a promising approach due to its decentralized structure and ability to preserve data locality. Yet, when applied to infrastructure systems, FL faces unique challenges that limit its current use. This project identifies three foundational barriers: ensuring robustness in adversarial environments, adapting to the hierarchical end-edge-cloud structure commonly found in infrastructure systems, and preserving privacy in the face of explainability and transparency requirements. The proposed research develops a comprehensive FL framework that addresses all three challenges in a unified manner. First, it explores new defenses and trust mechanisms that improve resilience against adversarial participants, guided by explainability of client behavior. Second, it introduces a hierarchical FL architecture that reflects real-world infrastructure deployment See more