Seminars

 

 

Prof. Dennis Akos (University of Colorado Boulder, USA)

Seminar title: Detecting and Localization of GNSS Radio Frequency Interference (RFI) - Jamming and Spoofing


Abstract: Global Navigation Satellite Systems (GPS/GNSS) have evolved into widely accessible, low-cost technologies, with receivers now available for approximately €1 (in volume) and capable of processing signals from multiple frequencies and constellations operated by the European Union, China, the United States, and Russia. As a result, modern society has become deeply reliant on these systems across a broad range of applications. It is estimated that a sustained disruption could cost the U.S. economy alone more than $1 billion per day. At the same time, the vulnerability of GPS/GNSS has become increasingly evident, particularly in the context of recent military conflicts, where interference and manipulation have been widely observed. This presentation provides an overview of satellite navigation systems, including their fundamental operating principles. It then examines the growing threat of radio-frequency interference (RFI), focusing on both jamming and spoofing. Finally, it outlines current research activities for detecting and localizing sources of such interference.

Go to Top

Prof. John Baras (University of Maryland, USA)

Seminar title: ML and AI methods for Networked Systems Security and Trust


Abstract: The explosive growth of ML and AI methods, algorithms and software, and the associated dramatic successes in several domains, have created the need to investigate their impact and consequences on security and trust. We present results on three specific domains. (i) Self-supervised time-series-anomaly detection with temporal logic explanations. The growing complexity of Cyber-Physical Systems (CPS) has heightened the need for robust, real-time anomaly detection to ensure safety and reliability in time-critical applications such as autonomous vehicles, industrial automation, etc. We present a novel framework for deep learning-based anomaly detection in CPS, specifically designed to operate on unlabeled multivariate time-series data. We integrate Signal Temporal Logic (STL) inference to provide meaningful, human understandable explanations for the decisions made. (ii) An End-to-End Encrypted Control Pipeline for Multi-Agent Coordination via CKKS Homomorphic Encryption. Cloud-based coordination of multi-agent systems requires sharing state with a central server, creating a conflict between coordination and privacy. Fully homomorphic encryption (FHE) resolves this in principle, but has severe arithmetic constraints. To ameliorate this problem we present an end-to-end encrypted control pipeline and open-source Julia implementation where sensing, state estimation, state propagation, and consensus control all operate on CKKS-encrypted data using only addition, multiplication, and cyclic rotation. The pipeline is validated on a multi-agent formation control scenario. (iii) ML and AI methods for Networked Systems Security and Trust. We first present results on the security of routing protocols for mobile adhoc networks (MANETs). We next introduce a rigorous foundation of AI emphasizing need to integrate ML with Knowledge Representation and Reasoning (KRR). We describe integration of LLMs and KGs towards efficient KRR systems. We describe results and open problems on the integration of ML, RL and KRR towards AI assistants, Agentic AI. We describe how these novel and emerging approaches hold promise towards efficient and high-performance AI assistants in security and trust.

Go to Top

Prof. Elisa Bertino (Purdue University, USA)

Seminar title: TBA


Abstract: TBA

Go to Top

Prof. Jean Camp (University of North Carolina at Charlotte, USA)

Seminar title: TBA


Abstract: TBA

Go to Top

Prof. Alexandra Dmitrienko (University of Duisburg-Essen, Germany)

Seminar title: Can We Trust AI? Understanding the Security Challenges of Modern Deep Learning Systems


Abstract: Recent advances in Deep Learning (DL) are transforming the way complex problems are addressed, enabling applications such as autonomous driving, medical and legal decision support, financial forecasting, and cybersecurity. At the same time, the increasing reliance on AI creates new opportunities for adversarial manipulation. Attackers may exploit vulnerabilities in DL systems to influence critical outcomes —for example, by causing self-driving cars to misinterpret traffic signs, manipulating financial predictions, or inducing incorrect medical or legal recommendations. Threats such as data poisoning, adversarial examples, and inference attacks challenge the integrity, confidentiality, and trustworthiness of AI systems. This lecture examines the security of DL systemsthroughout the model development and deployment lifecycle, covering common attack vectors, defense strategies, andhighlighting key challenges in designing resilient, trustworthy, and attack-resilient DL systems.

Go to Top

Prof. Amir Herzberg (University of Connecticut, USA)

Seminar1 title: It's time to upgrade the Public Key Infrastructure: Better Efficiency and Provable Security


Abstract: The PKI provides an essential foundation enabling public key cryptography, specified ~40 years ago and widely deployed, e.g., for web security, for ~30 years. It's time for a major upgrade; one reason is to ensure acceptable performance with PQC. It is also needed to improve security and privacy. We will present the challenges and solutions based on recent papers and ongoing standardization efforts. Based on joint works with: Jie Kong, Sara Wrótniak, Hemi Leibowitz and Ewa Syta and ongoing work in the IETF.

Seminar2 title: Deployable security for Internet routing security: State-of-the-Art


Abstract: Internet routing is notoriously vulnerable and abused for eavesdropping, MitM attacks, DDoS and more, in spite of extensive efforts. We explain the attacks and challenges , and discuss recent progress toward finally securing routing using cryptography. These include RPKI, BGPsec, ASPA and more, including several works by the author.

Go to Top

Dr. Beatrice Motella (European Commission, Joint Research Center, Italy)

Seminar title: GNSS Signal Authentication


Abstract: Global Navigation Satellite System (GNSS) spoofing and interference pose serious threats to any critical service that relies on accurate positioning, navigation and timing (PNT). In a spoofing attack, an adversary transmits counterfeit signals that mimic authentic GNSS broadcasts, causing receivers to compute false positions and potentially leading to hazardous outcomes. Interference, whether intentional or accidental, can weaken or block GNSS signals, resulting in loss of service or anomalous system behaviour.

At the system level, signal authentication is the primary defence, complemented by receiver‑side measures such as signal‑quality monitoring, adaptive antennas, machine‑learning‑based interference detection and the integration of external aiding navigation sources.

Among GNSS constellations, Galileo is the first to offer free civil authentication through the Open Service Navigation Message Authentication (OSNMA), which entered initial service in July 2025. OSNMA augments the Galileo Open Service by providing authentication data that confirm the integrity of navigation messages. The Galileo system is also developing the upcoming Signal Authentication Service (SAS), designed to further protect high‑priority applications.

The seminar will give an overview on GNSS vulnerabilities and the protection afforded by signal‑authentication techniques, with a particular focus on the OSNMA service offered by Galileo.

Go to Top

Prof. Catuscia Palamidessi (INRIA Saclay and LIX, France)

Seminar title: The EM Method for Private Frequency Estimation


Abstract: Differential privacy (DP) has become the gold standard for protecting sensitive information in individual-level data, spanning key variants like local DP and metric privacy. In this talk, we review standard privacy mechanisms, such as the Laplace and Gaussian mechanisms for DP and Randomized Response for Local DP, as well as the more sophisticated Blahut-Arimoto algorithm for metric privacy. A fundamental challenge remains: how to extract accurate aggregate information from data privatized via these mechanisms. This talk focuses on maximizing the utility of frequency estimates, a core primitive for downstream data analytics. Because Matrix Inversion (MI) is the most natural and direct approach for this task, it has been the first and most investigated method in the privacy literature. However, we champion a powerful but underutilized alternative from classic statistics: the Expectation-Maximization (EM) algorithm. We demonstrate that, while MI excels in computational speed, EM consistently provides superior accuracy when paired with metric privacy or deployed within federated learning environments. After contrasting the trade-offs between these two paradigms, we conclude with open questions and potential directions for future research.

Go to Top

Prof. Ahmad-Reza Sadeghi (TU Darmstadt, Germany)

Seminar title: Neuron Surgery for Machine Safety and Security


Abstract: Large Language Models are powerful, but they pose fundamental safety and security risks: they can be steered toward harmful outputs and have been shown to generate insecure or vulnerable code in realistic settings. Existing countermeasures combine external controls, such as filtering, guard models, and prompt engineering, with training-based alignment techniques, including Reinforcement Learning from Human Feedback and fine-tuning. While effective to a degree, these approaches primarily operate at the behavioral level, shaping outputs without explicitly controlling or understanding how safety and security are represented internally.

This talk argues for a shift in perspective: rather than focusing solely on external steering or global training objectives, we explore safety and security engineering from within the model itself. Emerging evidence suggests that safety alignment and secure code generation are, at least in part, mediated by small, specialized groups of neurons acting as internal control points.

Building on this insight, we explore a new paradigm: neuron-level surgery for machine safety and security. By directly targeting these critical neurons, we move beyond fine-tuning toward more precise and controllable interventions, with the potential to improve transparency and efficiency while enabling new forms of structural understanding, auditing, and control of model behavior. Finally, we outline key challenges toward realizing this vision.

Go to Top

Prof. Serge Vaudenay (EPFL, Switzerland)

Seminar title: Fair Exchange and Smart Contracts


Abstract: The fair exchange problem is to allow participants to exchange some assets in such a way that, under adversarial context, either all honest participants receive what they expect or no honest participant reveal their assets. This problem has been studied for 50 years. It is well known to be impossible to solve without any trusted third party. A natural idea is to use smart contracts over blockchains to solve some instances. In the case of an exchange of some crypto-currency against some algorithmically verifiable digital asset, several protocols exist. In this presentation, we survey some based on verifiable encryption and some based on interactive dispute resolution.

Go to Top

Prof. Ingrid Verbauwhede (KU Leuven, Belgium)

Seminar title: TBA


Abstract: TBA

Go to Top