Presentation Title:  Importance weighted variational inference without the reparameterization trick

Abstract. Modern AI relies on sophisticated inference methods to extract patterns from data while managing uncertainty. Importance Weighted Variational Inference (IWVI) is a powerful tool for this task, yet its development, analysis, and widespread adoption are largely restricted to settings where the so-called reparameterization trick holds. However, the reparameterization trick assumption prevents the use of IWVI in many challenging applications, such as those found in Bayesian phylogenetics and state-space modeling. In our work, we provide the first rigorous theoretical analysis of IWVI without this assumption. Our results expose fundamental deficiencies in existing methods and guide the development of new, theoretically sound, and computationally efficient algorithms whose effectiveness is demonstrated empirically.

Presentation Title:  HOPSIS - Hospitals' medical organization and Preparation to crisIS, ANR (2025-2028)

Abstract. The HOPSIS project aims to enhance hospitals' organizational ability to address emerging multiple risks by understanding the new forms of threats and crises faced by hospitals, and by analysing the current organization and preparation for major crises. It will specifically focus on the resilience of medical activity and the professional role of Crisis Medical Directors (hereafter CMD). Hospitals increasingly face significant events that threaten their structures, infrastructures, and the lives of employees and patients, while also managing exogenous crises such as a massive influx of patients following a major disaster, and health crises like COVID-19. In the face of such complex and hybrid crises, response efforts require greater agility, and adaptability. The project intends to:

1. Examine the new threats faced by hospitals and assess their impact on medical organizations

2. Better understand the career paths and trajectories of CMDs, and grasp their impact on decision-making, coordination, and preparation and their place in the wider hospital and healthcare system

3. Propose new training, preparation, and learning methods that integrate the professional insights and experiences of CMDs.

HOPSIS brings together three research groups from various disciplines: management, sociology, philosophy, political science, geopolitics, as well as healthcare researchers and practitioners. The project uses quantitative and qualitative methods, including a qualitative case studies of hospitals and a survey on medical crisis directors.

Presentation Title:  Intrafirm Knowledge Flows: How the Source of Knowledge and Status

Motives Shape Managers’ Knowledge Transfer Decisions

Abstract. Effective knowledge sharing across units is essential for multi-unit firms to realize synergies, foster innovation, and avoid duplication of effort. Yet such sharing is often impeded by internal frictions, particularly when managerial incentives and motives are misaligned. This study examines how managerial knowledge-sharing decisions are shaped by two key factors: the source of knowledge (internally generated vs. externally acquired) and the presence of status-based recognition from corporate headquarters. I conduct a context-rich experiment in which managers decide whether to share knowledge that was either created internally or acquired externally, with or without symbolic recognition from headquarters. In the absence of recognition, managers are more willing to share knowledge they created internally than knowledge acquired externally. However, this preference is significantly reduced when headquarters offers recognition for sharing, indicating that such status-oriented incentives can effectively shift managerial focus from individual credit to collective outcomes. These findings reveal an important asymmetry in how knowledge flows within firms and highlight the potential of symbolic recognition to reduce internal competition and foster collaboration. The study offers implications for the design of management control systems that leverage reputational motives, rather than relying solely on financial incentives, to overcome information frictions in complex organizational settings.

Presentation Title:  Globalization, Labor Market and Developing Countries, ANR (2025-2029)

Abstract. Over the past few decades, trade liberalization has played a central role in the structural reforms of developing countries. In recent times, considerable focus has shifted towards examining the microeconomic impacts of openness on labor markets and firms. The literature on the micro-economic effects of trade liberalization on labor markets have focused mainly on the formal manufacturing sector. This literature is not able to capture the impact of globalization on a large share of the labor force - almost half of economic activity - that is employed in the informal sector in the developing world. At the same time, the literature on firms and trade has so far largely neglected the key distinction between revenue productivity, i.e., the capacity of firms to generate revenue with a given amount of inputs and quantity productivity, i.e., the capacity of firms to generate physical output with a given amount of inputs. The project aims to contribute to filling these gaps.

Presentation Title:  A 28th Regime for European Innovative Companies

Abstract.The 2024 Draghi Report put forward numerous proposals to strengthen the European Union’s economic growth and capacity for innovation. Among them, the so-called ‘28th regime’, which entails the creation of a fully harmonised corporate form at the European level, proved especially popular. This paper questions the merits and feasibility of this project. After outlining the two prevailing conceptions of the 28th regime, it identifies three key regulatory challenges that European innovative companies face from a corporate law perspective: administrative burdens, rigid mandatory rules, and legal fragmentation. The paper then examines how these challenges can best be addressed. It argues that the establishment of a fully fledged European corporate form would encounter substantial political and technical barriers, and demonstrates that these problems could be resolved just as effectively through a series of targeted reforms benefiting existing national corporate for

Presentation Title:  BIPDM – Analyses comportementales dans la prise de décision liée aux PRIIPs, CY Initiative – Programme EMERGENCE (2025-2027)

Abstract. Investor decision-making often deviates from optimal strategies due to behavioral biases and cognitive limitations, such as the disposition effect, overconfidence, limited attention, and information processing errors. These factors lead to suboptimal financial outcomes, which can significantly erode long-term returns. Regulatory measures like the UCITS KIID and PRIIPs KID aim to enhance transparency and empower investors, yet aligning investor behavior with cost-efficient decision-making remains a challenge. This research examines the impact of the PRIIPs KID’s innovative disclosure features on investor behavior. The study explores how investors allocate cognitive resources, respond to scenarios, and process detailed cost information, while also examining the roles of financial literacy and skewness preferences. Using real-world PRIIPs KIDs in a controlled experimental setting, participants will make investment decisions, enabling the study to analyze behavioral factors driving choices. Testable hypotheses address key challenges, including cognitive overload and limited attention.

The research contributes to behavioral finance and regulatory policy by evaluating the effectiveness of enhanced disclosure formats and advancing understanding of how document design influences decision-making. It also sheds light on the effects of document complexity, cognitive limitations, and biases, offering actionable insights for policymakers and financial institutions to optimize disclosures, enhance investor outcomes, and support better decision-making.

Presentation Title:  The distributional effects of low emission zones: Who benefits from cleaner air?

Abstract. Low emission zones (LEZ) represent a key environmental policy instrument to address air pollution in cities. LEZs have reduced air pollution and associated health damages in regulated areas, but it remains unclear who has benefited from cleaner air. To examine the distributional effects of LEZs, we combine gridded data on resident characteristics, including income and a proxy for ethnicity, with high-resolution estimates of fine particle (PM2.5) concentrations in Germany, the country with the largest number of LEZs. We estimate heterogeneous treatment effects with a difference-in-differences approach and show that PM2.5 pollution reductions are distributed unequally across society. While residents with German name origins experience larger improvements within LEZs, residents with foreign names disproportionately live in LEZs and thus benefit more when assessed at a nationwide scale. Monetizing air quality benefits following governmental guidance, we find that they are distributed pro-poor within LEZs, disproportionately benefiting lower-income residents. From a nationwide perspective, however, benefits are distributed almost proportionally, while the sign is sensitive to how benefits from cleaner air scale with income. Overall, our results suggest that LEZs have nuanced distributional implications that differ sharply between a national perspective and local assessments that focus on effects within LEZs.

Presentation Title:  RIVERS – Revise Intergenerational Values for Sustainability, CY Generation – HORIZON (2025–2028)

Abstract. This project investigates the design and survival of systemic S-frame incentives, to secure intergenerational adoption of sustainable practices by organizations. We implement an interdisciplinary approach to examine the way in which institutions are transformed, while maintaining a sense of intra and intergenerational connection, in which values such as cooperation, altruism, diversity and equity are supported through legitimate incentives, and transform into organizational standards. Protecting natural resources, but also diversity of social capital, as well as equality for future generations (which we all consider as “public goods”) are imperatives of sustainability and rationales for costly societal efforts today. We believe that by using appropriate visualizations, data physicalization (such as materialized numbers and graphs about the quality of the air, the state of inequalities, the diversity standards or a mapping of CSR practices...) and simulations of the many states of the world, we can design controlled experiments in which intra and intergenerational links can be considered together, and perceived globally, which will be one of the major methodological novelty of our project. Exploring the use of better visualizations about norms, regulations, routines, institution design adoption, their diffusion, evolution and, at the end, survival and improvement, will help overcoming memory shortness and incomplete backward and forward perspective and data availability by mapping, in an intuitive way, past behavior and data, and making it salient. This design will allow strategic computations about joint norms adoption and modelization of their interactions. 

Presentation Title:  LEARN20PT – Learning to Optimize : A Sustainable Approach, CY Initiative – HORIZON (2025–2027)

Abstract. The project aims to develop a novel framework that integrates machine learning (ML) with classical optimization to solve complex problems in real-time within decentralized environments. It addresses the trade-off between solution exactness and speed, crucial for real-time applications where traditional exact methods are computationally expensive and time-intensive. Using supervised and reinforcement learning, the framework approximates optimal solutions by learning from previously solved problems, enabling efficient problem-solving under tight time constraints. A significant challenge tackled by the project is ensuring privacy in decentralized systems, critical for fields like healthcare, finance, and supply chain management. Sensitive data often cannot be centralized due to privacy concerns. The proposed decentralized approach allows subproblems to be solved locally on different network nodes while ensuring data privacy and enabling scalability. Splitting problems into smaller, parallelizable subproblems significantly reduces overall solution time. The framework leverages decomposition techniques like column generation and Benders decomposition, breaking large problems into a master problem and smaller, manageable subproblems. In a decentralized setup, each node solves subproblems independently, using ML-enhanced optimization. Machine learning models, trained on historical optimization data, predict decision variables or constraints, delivering near-optimal solutions faster than solving subproblems from scratch. By integrating ML with decomposition, the system achieves faster problem-solving and adapts to new instances over time. Continuous learning ensures improved efficiency, eliminating the need for exhaustive search algorithms and making the framework ideal for real-time decision-making. This approach combines speed, scalability, and privacy, providing a robust solution for large-scale optimization challenges.

Presentation Title:  ARC Centre of Excellence for Quality Work in a Digital Age (QWIDA)

Funding Body: Australian Research Council (ARC) 

Abstract. The QWIDA Centre aims to bring together leading experts from the social and technical sciences to shape the future of quality work in the era of AI and automation. By harnessing emerging technologies, the Centre will develop practical tools and actionable insights to enhance human performance, foster collaboration across boundaries, and equip workers with the skills needed to thrive in a digital world. In doing so, it will contribute to improved wellbeing, inclusion, and productivity across both government and industry. Project stakeholders are Sharon K. Parker, the director of the Centre, at Australia’s Curtin University, and Lauren Waardenburg, the ESSEC representative.