Upcoming Seminars

Abstract: Encrypted control offers a promising solution for enhancing security in networked control systems, enabling direct control operations over encrypted data. However, its implementation is often challenged by the constraints of homomorphic encryption, particularly the requirement that linear dynamic controllers should ``have integer coefficients.’’ This talk explores the rationale behind this requirement and addresses the problem of designing such controllers. Specifically, I will present a recent finding that a stabilizing controller with integer coefficients always exists for any given discrete-time linear time-invariant (LTI) plant, along with a constructive method to find one. Furthermore, I will discuss the problem of converting a pre-designed controller into one with integer coefficients, while preserving the original performance.

Abstract: Federated Learning (FL) is widely viewed as a privacy-preserving paradigm because raw training data never leaves client devices. However, this data-centric perspective overlooks a crucial and largely unprotected asset: the model itself. In many practical FL deployments, client updates are transmitted over networks that may be monitored by passive eavesdroppers. Such adversaries can aggregate intercepted updates over time to reconstruct high-quality surrogate models, posing serious risks to intellectual property, system integrity, and downstream security. This talk shifts the focus of FL security from data confidentiality to model confidentiality. We develop a theoretical framework to analyze how factors such as client sampling probability, local objective structure, server aggregation, and adversary capabilities influence the extent to which an eavesdropper can reconstruct the global model. Our analysis reveals inherent vulnerabilities in standard FL protocols and highlights limitations of differential privacy when used to protect model parameters rather than data. We then present a lightweight, architecture-agnostic defense based on dynamic uniform quantization, repurposed specifically for model protection. We show that this approach provides provable, persistent protection against passive reconstruction attacks. Overall, this work frames model confidentiality as a core security problem in distributed learning systems.

Abstract: Positioning and tracking are key enablers of a wide range of applications that require reliable and accurate location information. However, these technologies are vulnerable to both intentional attacks and unintentional interference, motivating the need for resilient solutions. This talk presents probabilistic, uncertainty-aware approaches to two fundamental challenges that threaten situational awareness in today’s hostile environments. First, we study resilient satellite-based navigation under infrastructure outages. We propose cooperative positioning strategies in large-scale real-time kinematic networks that achieve centimeter-level accuracy despite missing or mixed-quality reference data, and analyze performance as a function of network size, geometry, and robustness to outliers from jamming attacks and multipath. Second, we counter deception jamming in radar-based localization using multi-target tracking (MTT) frameworks based on random finite set theory. In particular, we focus on range gate pull-off attacks, which generate adversarial radar returns intended to deceive the tracker into following false targets. By exploiting attack characteristics within the MTT algorithm, we significantly reduce spoofed track persistence. Together, these contributions provide a path toward resilient navigation and sensing systems through probabilistic and Bayesian methods.