Dear All,
The next seminar(s) will take place on 14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany) - D1_Room_0.15. Presenters and their advisors are encouraged to present in person. We especially encourage other students and teachers to attend and present in person as well.
For presenters,
1. We would book the room half an hour in advance, so you are encouraged to arrive a few minutes early to set up your own poster.
2. For this session, you need to print the poster on your own. The size of the poster should be 116x86cm or 86x116cm. You can use the poster printing service of Saarland University (https://www.uni-saarland.de/en/page/uds-card/functions/printing.html -> Posterdruck A0).
3. You need to present your poster in a much smaller group, but you are encouraged to roam around and ask questions about other posters.
4. We encourage you to bring your laptop to present your demo; there will be small tables in the room where you can put your laptop.
Presenters: Jonas Schmitt, Muhammad Saqib, Nomaan Shaikh, Sumair Ijaz Hashmi, Mutahar Aamir, Adham Mahgoub, Ved Rahul Shinkre, Aditya Gupta, Oleksii Palagin, Anshul Birwal
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Jonas Schmitt
Type of Poster: Master Intro
Advisor: Jonas Hielscher
Title: Exploring the Gap Between Human-Centered Security Research and Organizational Practice
Research Area: RA6: Empirical and Behavioural Security
Abstract: Human-Centered security (HCS) shifts the focus from pure computational problems to the users. Information security within companies is therefore also dependent on usability and employee focused organizational security. Many companies use phishing simulations and unified awareness training to educate their employees about the protection of corporate data. In research literature on the other hand we observe a shift towards individual threat evaluation and training. This thesis explores the gap between research and organizational practice from an educational perspective. Previous work has questioned the effectiveness of widely used organizational security practices such as phishing simulations. Interviews with chief information security officers (CISOs) then found that reasons for this gap include the lack of familiarity with the current research literature as well as company specific compliance rules. This work aims to find out whether the next generation of security actors learn the relevant competencies in their higher education to bridge this gap. To do so, we present a qualitative research approach that uses semi-structured interviews with HCS educators around the world to determine teaching focuses, importance in university curricula and challenges that they face. The collected data will be assessed via thematic analysis to extract overarching themes and patterns. With this study we aim to understand the representation of HCS in higher education and how educators perceive its role and recognition. This gives us grounds for predictions on the HCS knowledge the next generation of security actors have prepared for their future employers. We are going to develop guidelines for curricula designs that provide a stronger focus on the human factor.
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Muhammad Saqib
Type of Poster: Master Intro
Advisor: Mario Fritz, Sarath Sivaprasad
Title: A Behavioural and Representational Investigation of Graph World Models in Open-Weight Reasoning Models
Research Area: RA2: Trustworthy Information Processing
Abstract: This thesis aims to investigate whether open-weight reasoning models develop and utilize limited, task-specific world models for graph-based planning. It compares internal representations formed while models observe random walks with the representations that evolve during self-generated reasoning trails. Using unseen graph instances i.e., a controlled subset of CogEval-inspired tasks, the models will perform route planning and adapt to detours, shortcuts and goal changes. Behavioural evaluation will mainly measure route validity and adaptation while token-level analysis will examine how graph structure develops across layers and the reasoning steps. Probes and targeted causal interventions will distinguish information that is merely decodable from representations that models actively update and utilize for planning. The thesis will provide evidence for or against graph world models in open-weight reasoning models while avoiding broader claims about human-like world understanding.
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Nomaan Shaikh
Type of Poster: Master Intro
Advisor: Andrea Mengascini
Title: Improving Usability in Security Dashboards: Designing and Evaluating an Empowered Product Team-Centric Interface for Security KPIs and Findings Management
Research Area: RA6: Empirical and Behavioural Security
Abstract: Modern day software development practices integrate security responsibilities that are to be managed by Empowered Product Teams (a.k.a EPTs). As an organization adopts DevSecOps, and cloud-native development models, developers are required to assess, interpret, and remediate security findings whilst their daily engineering tasks. To support this shift, organizations often use dashboards to visualize and provide information about their products active findings during the software life cycle. However, these dashboards are often designed with the assumption and mental model of security experts, creating significant usability challenges for developers. From high cognitive load, inconsistent terminology and metrics, to poor visualization of crucial data. These elements prevent EPTs from understanding vulnerabilities that require immediate attention or how these findings are related to their system. In this thesis we explore and perform usability studies on users to understand the pain points of the current security dashboard, that brings data from multiple different sources, and visualize them into one central dashboard.
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Sumair Ijaz Hashmi
Type of Poster: Master Intro
Advisor: Katharina Krombholz
Title: Understanding the Threat Landscape of Low Socioeconomic Pakistanis
Research Area: RA6: Empirical and Behavioural Security
Abstract: Online cybersecurity and privacy technology is typically designed with literate and Westernized users in mind. Hence, smartphone users from low socioeconomic backgrounds, especially those who are illiterate, face severe security challenges when navigating the online world. However, little is known about how low socioeconomic status (LSES) smartphone users experience, perceive, and defend against threats under financial, literacy, and social constraints. This thesis aims to explore the threat landscape of LSES smartphone users in Pakistan to understand which threats and harms they face, how they perceive risk, and how they protect themselves and each other.
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Mutahar Aamir
Type of Poster: Master Intro
Advisor: Rebekka Burkholz
Title: Post-hoc Pruning with Gradients and Resampling
Research Area: RA2: Trustworthy Information Processing
Abstract: We aim to increase the efficiency of Large Language Models by reducing their parameter count via pruning. Static masks are found through small calibration datasets, but model quality collapses at higher sparsities. Dynamic sparse training fixes bad masks, but requires multiple rounds of retraining, which is computationally infeasible for LLMs. We propose a new technique inspired by dynamic sparse training that uses gradient information as well as resampling from the calibration dataset to find the optimal mask. We generalize this technique so that it can be applied on top of existing masks, leading to more robust sparse models
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Adham Mahgoub
Type of Poster: Master Intro
Advisor: Pouya Narimani, Andreas Zeller, Ali Abbasi
Title: Software Composition Analysis using Power Side-channels
Research Area: RA5: Secure Mobile and Autonomous Systems
Abstract: Identifying software components running on embedded devices is challenging when the devices can only be analyzed as black boxes, without access to their firmware binaries or source code. This lack of access complicates vulnerability assessment because embedded firmware commonly incorporates third-party open-source libraries that may remain outdated long after known vulnerabilities have been patched. Therefore, identifying the libraries used by a device can help assess its exposure to known vulnerabilities. Power side-channels provide a potential information source for such identification. The power consumption of a microcontroller reflects aspects of its internal code execution and can be observed without direct access to the firmware. In this work, we investigate whether power side-channel measurements can be used to fingerprint library functions from commonly used open-source libraries and subsequently identify their execution within an unknown firmware. Our evaluation studies the feasibility and practical requirements of power side-channel-based software composition analysis for black-box embedded systems by examining the extent to which power traces can be segmented into regions associated with different library functionalities, the robustness of identification across devices, and the minimum measurement capabilities required for reliable detection.
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Ved Rahul Shinkre
Type of Poster: Bachelor Intro
Advisor: Mario Fritz, Philipp Christmann, Tejumade Afonja
Title: Differentially Private Multimodal Tabular Data Generation
Research Area: RA2: Trustworthy Information Processing
Abstract: Generating synthetic tabular data becomes difficult when datasets contain both structured attributes and free-form text. Current generative approaches typically specialise in only one modality: tabular models accurately capture numerical distributions and categorical dependencies, but cannot handle unstructured text. Conversely, large language models produce fluent and realistic text, but struggle to maintain exact numerical correlations, statistical distributions, and mathematical consistency. To address this trade-off, this thesis proposes a three-stage generative pipeline designed to balance both modalities without sacrificing quality. First, unstructured text is compressed into tabular features by extracting domain-specific entities, numerical values, and stylistic properties. Second, tabular generative models learn and synthesise the joint distributions across these extracted attributes, ensuring that mathematical and statistical relationships are properly preserved. Finally, a language model conditioned on the synthetic tabular features re-inflates the attributes back into coherent natural language text. By separating statistical distribution modelling from text generation, this framework aims to generate realistic synthetic data that preserves both strict numerical integrity and high-quality text.
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Aditya Gupta
Type of Poster: Master Intro
Advisor: Katharina Krombholz, Simon Anell
Title: OPEN-SOURCE DEVELOPERS’ REPOSITORY TRANSFER STRATEGIES
Research Area: RA6: Empirical and Behavioural Security
Abstract: Open-source software plays a crucial role in technological innovation and research. A 2024 Harvard Business School study estimates that without open-source software, firms would spend more than 3.5 times what they currently spend on software. This widespread dependence creates significant security risks. Supply chain attacks targeting open-source projects could affect numerous downstream dependents. Existing research has focused primarily on the technical perspectives of such attacks, and less attention has been given to the human and organisational factors involved in maintaining and transferring the responsibilities of such repositories. This study investigates how open-source developers perceive and manage these security risks. Following a qualitative approach, this study aims to identify the challenges and practices surrounding repository responsibilities, as well as the roles and support required from the repository platforms and the broader open-source community.
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Oleksii Palagin
Type of Poster: Master Intro
Advisor: Wouter Lueks, Sylvain Chatel
Title: PIVO: Privacy-Preserving Indexing and Verification of Record Overlaps
Research Area: RA2: Trustworthy Information Processing
Abstract: Record linkage identifies records referring to the same person across databases, even when identifying attributes contain errors or missing values. In privacy-sensitive settings such as healthcare, secure multi-party computation enables this process without disclosing raw identifiers. However, its computational and communication costs remain a significant scalability challenge. This thesis presents PIVO, an integrated system for privacy-preserving record linkage that addresses both the number of comparisons and the cost of secure matching. PIVO combines locality-sensitive hashing and private filtering with a purpose-built three-party MPC library. The library uses packed data representations, batched operations, and specialized routines for similarity computation and secure best-match selection. Together, these components reduce unnecessary comparisons and optimize the remaining matching work. Weighted fuzzy and exact comparisons, missing-field handling, and final classification are evaluated inside MPC. The privacy-oriented configuration additionally hides intermediate candidate relationships and individual filtering decisions under a semi-honest, honest-majority model. The contribution is the design, implementation, and evaluation of these techniques within one configurable linkage system.
14.10.2026, 14:00 - 16:00, CISPA D1 (Kaiserstraße 21 66386 St. Ingbert, Germany)
Presenter: Anshul Birwal
Type of Poster: Bachelor Intro
Advisor: Sajin Sasy
Title: Scaling Metadata-Protecting Communications using PIR & Intersecting Topologies
Research Area: RA1: Algorithmic Foundations and Cryptography
Abstract: Everyday messaging apps like WhatsApp, Signal etc. use end-to-end encryption. E2E protects the message content but leaves the sensitive communication metadata exposed. To protect this metadata, some existing systems like Pung rely on single-server PIR. Senders deposit encrypted messages into a central database, and recipients retrieve them without revealing the accessed index, resulting in computational cost for each read to be $O(n)$ for server. In private-write based systems like Express, users insert their message into a particular recipient's mailbox in database without letting the server know whose mailbox it was. This also results in $O(n)$ computation cost for the server. In this thesis, we propose a new architecture that breaks this bottleneck by mapping the intersecting routing topologies traditionally used in multi-server mix-nets onto a single-server PIR database. We structure the database geometrically using Finite Projective Planes (FPP), where each user is assigned a static subset of $O(\sqrt{n})$ mailbox slots. The algebraic properties of FPP guarantee that any two users’ subsets intersect at exactly one unique mailbox.
