ProPILE: Probing Privacy Leakage in Large Language Models

in Advances in Neural Information Processing Systems 36 (NeurIPS 2023) (2023, December)

The rapid advancement and widespread use of large language models (LLMs) have raised significant concerns regarding the potential leakage of personally identifiable information (PII). These models are ... [more ▼]

The rapid advancement and widespread use of large language models (LLMs) have raised significant concerns regarding the potential leakage of personally identifiable information (PII). These models are often trained on vast quantities of web-collected data, which may inadvertently include sensitive personal data. This paper presents ProPILE, a novel probing tool designed to empower data subjects, or the owners of the PII, with awareness of potential PII leakage in LLM-based services. ProPILE lets data subjects formulate prompts based on their own PII to evaluate the level of privacy intrusion in LLMs. We demonstrate its application on the OPT-1.3B model trained on the publicly available Pile dataset. We show how hypothetical data subjects may assess the likelihood of their PII being included in the Pile dataset being revealed. ProPILE can also be leveraged by LLM service providers to effectively evaluate their own levels of PII leakage with more powerful prompts specifically tuned for their in-house models. This tool represents a pioneering step towards empowering the data subjects for their awareness and control over their own data on the web. [less ▲]

Luxembourg. ECEC Workforce profile

Report (2023)

A survey on deep learning-based monocular spacecraft pose estimation: Current state, limitations and prospects

in Acta Astronautica (2023), 212

Estimating the pose of an uncooperative spacecraft is an important computer vision problem for enabling the deployment of automatic vision-based systems in orbit, with applications ranging from on-orbit ... [more ▼]

Estimating the pose of an uncooperative spacecraft is an important computer vision problem for enabling the deployment of automatic vision-based systems in orbit, with applications ranging from on-orbit servicing to space debris removal. Following the general trend in computer vision, more and more works have been focusing on leveraging Deep Learning (DL) methods to address this problem. However and despite promising research-stage results, major challenges preventing the use of such methods in real-life missions still stand in the way. In particular, the deployment of such computation-intensive algorithms is still under-investigated, while the performance drop when training on synthetic and testing on real images remains to mitigate. The primary goal of this survey is to describe the current DL-based methods for spacecraft pose estimation in a comprehensive manner. The secondary goal is to help define the limitations towards the effective deployment of DL-based spacecraft pose estimation solutions for reliable autonomous vision-based applications. To this end, the survey first summarises the existing algorithms according to two approaches: hybrid modular pipelines and direct end-to-end regression methods. A comparison of algorithms is presented not only in terms of pose accuracy but also with a focus on network architectures and models' sizes keeping potential deployment in mind. Then, current monocular spacecraft pose estimation datasets used to train and test these methods are discussed. The data generation methods: simulators and testbeds, the domain gap and the performance drop between synthetically generated and lab/space collected images and the potential solutions are also discussed. Finally, the paper presents open research questions and future directions in the field, drawing parallels with other computer vision applications. [less ▲]

Workshop: An Introduction to the Theory and Methods of Data Driven Research

Diverse speeches and writings (2023)

Workshop for the EUROPAST project Lund workshop.

Maschinelles Lernen in der automatischen Auswertung von Punktwolken: Beispiele der Erfahrungen an der Universität Luxemburg

Scientific Conference (2023, October 07)

Challenges in completing EMU: asymmetric competition vs fiscal harmonisation. A case study of the Benelux countries

in Journal of Contemporary European Research (2023)

This paper aims to investigate the concept, context and socio-economic consequences of fiscal competition in the integrated economic space of EMU in completion, to pinpoint the positive and negative ... [more ▼]

This paper aims to investigate the concept, context and socio-economic consequences of fiscal competition in the integrated economic space of EMU in completion, to pinpoint the positive and negative factors at work via a case study of the Benelux countries – both founder members of the EU and pioneers of EMU – and to examine the impact on European and international regulations in the field. In particular, it will endeavour to provide a comprehensive interpretation of fiscal policy in the Benelux countries via a comparative approach and from a historical perspective. It will look at the development of respective domestic fiscal policies, driven by national interests and by membership of a Community that is subject to requirements in terms of harmonisation and taxation, but also by constant contact (and frequent clashes) with the multilateral international environment. [less ▲]

Transkultureller Theaterschauplatz: Grenzen und die Odyssee Fliehender – Interdisziplinäre Überlegungen

Book published by Melusa Press (2023)

Der vorliegende Sammelband versammelt kritische Forschungsansätze sowohl aus verschiedenen akademischen Disziplinen – der Theater-, Medien , Film-, Literaturwissenschaften sowie der Soziologie, der ... [more ▼]

Der vorliegende Sammelband versammelt kritische Forschungsansätze sowohl aus verschiedenen akademischen Disziplinen – der Theater-, Medien , Film-, Literaturwissenschaften sowie der Soziologie, der Erziehungswissenschaften – als auch der praxisbezogenen Theaterpädagogik und der Sozialen Arbeit. Anliegen ist es, die interdisziplinäre Forschung und einen integrativen Forschungsansatz für den Sammelband fruchtbar zu machen, um so den gegenseitigen akademischen und praktischen Transfer von Wissen hervorzuheben und die Kluft zwischen Wissen und Praxis in Bezug auf Wissenschaft, Theater und Migration zu überwinden: Dabei geht es um die Analyse des komplexen Themenfeldes, das die gegenwärtigen Migrationsbewegungen ins Zentrum der internationalen und interdisziplinären Diskussionen rückt. [less ▲]

PROACTIVE COMPUTING PARADIGM APPLIED TO THE PROGRAMMING OF ROBOTIC SYSTEMS

Doctoral thesis (2023)

This doctoral thesis is concerned with the development of advanced software for robotic systems, an area still in its experimental infancy, lacking essential methodologies from generic software ... [more ▼]

This doctoral thesis is concerned with the development of advanced software for robotic systems, an area still in its experimental infancy, lacking essential methodologies from generic software engineering. A significant challenge within this domain is the absence of a well-established separation of concerns from the design phase. This deficiency is exemplified by Navigation 2, a realworld reference application for (semi-) autonomous robot journeys developed for and on top of the Robot Operating System (ROS): the project’s leading researchers encountered difficulties in maintaining and evolving their complex software, even for supposed-to-be straightforward new functions, leading to a halt in further development. In response, this thesis first presents an alternative design and implementation approach that not only rectifies the issues but also elevates the programming level of consistent robot behaviors. By leveraging the proactive computing paradigm, our dedicated software engineering model provides programmers with enhanced code extension, reusability and maintenance capabilities. Furthermore, a key advantage of the model lies in its dynamic adaptability via on-the-fly strategy change in decision-making. Second, in order to provide a comprehensive evaluation of the two systems, an exhaustive comparative study between Navigation 2 and the same application implemented along the lines of our model, is conducted. This study covers thorough assessments at both compile-time and runtime. Software metrics such as coupling, lack of cohesion, complexity, and various size measures are employed to quantify and visualize code quality and efficiency attributes. The CodeMR software tool aids in visualizing these metrics, while runtime analysis involves monitoring CPU and memory usage through the Datadog monitoring software. Preliminary findings indicate that our implementation either matches or surpasses Navigation 2’s performance while simultaneously enhancing code structure and simplifying modifications and extensions of the code base. [less ▲]

Blockchains, AI, and the uncertain promise of calculated certainties

Presentation (2023, September 27)

Every culture chooses its own approach and creates its own authority to provide answers to the uncertainties of human life in socety. Since God has become obsolete as the guarantor of our society, it is ... [more ▼]

Every culture chooses its own approach and creates its own authority to provide answers to the uncertainties of human life in socety. Since God has become obsolete as the guarantor of our society, it is up to instances such as the State, the People or the Market and the corresponding institutions to act in a way that creates meaning and reduces uncertainty. In the face of worsening crises (climate, wars, representation...), however, these instances are also increasingly losing their binding power, effectiveness and legitimacy. Increasingly, therefore, the focus lies on algorithmic-based technologies, which, due to their number-based approach, are seen as the more reliable supplier of answers to the most pressing questions of our time. Unlike in the case of the previously mentioned instances, here, at least it seems, belief in the validity of their statements is no longer necessary: rather, by resorting to ever more extensive amounts of data, social and individual life is calculated, predictable and correspondingly controllable. Moreover, in the same way - in the sense of technological solutionism - it should be possible to defuse the most pressing crises.Two technologies in particular come to the fore: With blockchain technology, the last factor of uncertainty - the middleman legitimising transactions - is now eliminated: Via algorithmically controlled consensus and automation processes, the decentralised, transparent and unforgeable control and regulation of binary and social processes will become possible and a "more reliable truth than any truth we have ever seen" (Casey/Vigna) will be created. With (General) Artificial Intelligence, an entity is supposed to emerge that, on the basis of extensive data and self-learning algorithms, can replace the human, i.e. fallible intellect, and - eradicating any uncertainties - calculate and answer the unanswerable questions of human life. Against the backdrop of the recent French history of ideas, I will explore in my paper to what extent this is actually a new - safer - form of uncertainty reduction or not; in other words, I will pose the question to what extent blockchains and AI actually eliminate uncertainties or exacerbate existing uncertainties or create entirely new uncertainties. [less ▲]

Don’t try to be original! Web archives as ‘reborn digital sources’

Scientific Conference (2023, September 21)

Web archives as “born digital” source collections come with multiple biases, both concerning the archived web pages and web sites themselves, the web collections (notably those created on precise events ... [more ▼]

Web archives as “born digital” source collections come with multiple biases, both concerning the archived web pages and web sites themselves, the web collections (notably those created on precise events, like Paris terrorist attacks and the COVID-crisis), and the metadata. After explaining why web archives must be considered as a “reborn digital heritage” (Brügger,2016) and the multi-layered socio-technical transformations and mediations that are inherent to the archiving of the Web, the presentation will focus on scholary strategies to face such uncertainties, including best practices in documentation, scalable reading, and online source criticism. [less ▲]