An important problem in machine learning and computational statistics is to sample from an intractable target distribution. In Bayesian inference for instance, the latter corresponds to the posterior distribution of the parameters, which is known only up to an intractable normalisation constant, and is needed for predictive inference. In deep learning, optimizing the parameters of a big neural network can be seen as the search for an optimal distribution over the parameters of the network. This sampling problem can be cast as the optimization of a dissimilarity (the loss) functional, over the space of probability measures. As in optimization, a natural idea is to start from an initial distribution and apply a descent scheme for this problem. In particular, one can leverage the geometry of Optimal transport and consider Wasserstein gradient flows, that find continuous path of probability distributions decreasing the loss functional. Different algorithms to approximate the target distribution result from the choice of a loss functional, a time and space discretization; and results in practice to the simulation of interacting particle systems. This optimization point of view has recently led to new algorithms for sampling, but has also shed light on the analysis of existing schemes in Bayesian inference or neural networks optimization. However, many theoretical and practical aspects of these approaches remain unclear. First, their non asymptotic properties quantifying the quality of the approximate distribution at a finite time and for a finite number of particles. Second, their convergence in the case where the target is not log-concave (which is analog to the non-convex optimization setting). Motivated by the machine learning applications mentioned above, the goal of this project is to investigate these questions, by leveraging recent techniques from the optimization, optimal transport, and partial differential equations literature.

Over the last decades, internet has sped up and increased interactions between employers and workers, but aggregate unemployment does not seem to have been much impacted by this revolution. This could be because information frictions are not a first-order contributor of unemployment, or because current tools and institutions do not enable truthful and effective communication between firms and workers. Employers, who are often on the short side of the market, find it difficult and costly to screen potential employees. INASHI aims to provide theoretical frameworks and new empirical evidence about what the remaining information imperfections on the labour market are, how important they are to aggregate unemployment and unemployment of the most vulnerable segments of the labour market, and what solutions can be put in place to improve the recruiting process. INASHI will combine novel data on how firms search for workers on large online job boards with administrative data on vacancies, and matched employer-employee data. It will also leverage a series of randomised controlled trials to test how the provision of new information to employers, whether about candidates or about features of the market, help them make better hiring decisions, leading ultimately to higher aggregate hiring, and higher-quality matches. Three countries will be studied, Austria, France, and Sweden, so that INASHI will provide evidence valid in a variety of contexts.

Decades of empirical research in behavioral economics and psychology have generated abundant evidence showing that individuals’ decisions in economically relevant situations often deviate systematically from the paradigm of perfect rationality. Two categories of anomalies have garnered particular attention. The first category groups belief biases and learning distortions, primarily focusing on biases in retrospective learning, that is, distortions in the process through which individuals interpret and use a piece of information that has already been revealed. The second category documents anomalies in time preferences and discounting behavior, with a specific emphasis on the phenomenon of time inconsistency and self-control issues. The purpose of this project is to redirect attention toward a facet of our cognitive life that is of paramount importance in dynamic decision problems yet has received little attention from economists thus far: the forecasting of the usefulness of a piece of information before its revelation. This type of forecast conditions important decisions, such as how much information to acquire, how much to experiment, or how much to pay attention to information sources. This project aims to evaluate individuals’ ability for prospective reasoning in economically relevant situations. Our objective is to provide evidence for the following questions: Do people experiment as much as they should in their daily lives? Do they gather enough information before making important life decisions, such as such as college applications, financial investments, medical decisions, or consumption choices? Or do they miss opportunities to make better choices because they fail to appreciate the value of active learning and experimentation?

The distribution of workers across jobs and across different geographic areas has major implications for growth, social welfare and inequality. It is complex but essential to understand the mechanisms behind the allocation of talents in the economy, that is, how skilled individuals choose their jobs and where they live. However, the existing literature on the distribution of skills faces two major challenges. First, it has largely neglected the role of the marriage market and family constraints, although family formation and partner choice are intimately linked to career and location choices and these choices influence each other. Second, it must go beyond the standard one-dimensional classification of skills based on educational attainment alone. This hierarchy underestimates the inequalities that exist between multidimensional and non-hierarchical skill sets. SkiM2Lab will address this challenge in developing state-of-the-art multidimensional matching models with two specific objectives. In the first objective, I will analyze the interactions between the labor market and the marriage market using equilibrium models of matching where individuals and jobs are associated with multidimensional skill sets and are located in different places. Estimating these structural models on household data will reveal how family and labor markets affect wage disparities and occupational segregation by gender and region. In the second objective, I will leverage big data such as online resumes and online job postings, as well as machine learning and natural language processing technologies, to extract skills at the most granular level, build new relevant combinations of skills and include them in a competitive matching model. This will make it possible to propose a new method to identify rapidly relevant emerging skills and their impact on wages and production.

The gender revolution framework predicts a seemingly linear progression leading to a dramatic convergence in men’s and women’s roles in paid work and at home. Yet gender convergence appears stalled by conflicting structural and cultural factors across industrialized countries. Existing theoretical perspectives fail to simultaneously predict how the gender revolution shapes couple-level work-family patterns across countries and time for those with lower, middle, and higher socio-economic resources. WeEqualize will address the intertwined implications of the gender revolution—including changing gender beliefs, rising labor market insecurity, and the increasing retreat from partnerships—in shaping social inequalities in work-family strategies among different-sex couples across 24 countries from the 1960s to nowadays. WeEqualize will provide the first comprehensive characterization and quantification of social inequalities in work-family strategies across industrialized countries and over the long run. It aims to: identify a couple-level typology of work-family strategies; examine the prevalence of these strategies by education and across countries; evaluate the role of contextual factors in shaping work-family strategies; assess how historical and contemporary estimates of work-family strategies are shaped by changing demographic trends, and project future trends in work-family strategies for the coming decades; as well as collect and leverage new survey-based experimental data across different contexts to disentangle the role of gender beliefs from labor market constraints in shaping what type of work-family strategies couples choose and why. By combining innovative computational methods with multiple nationally representative studies, as well as collecting new survey-experimental data, WeEqualize will challenge and reframe our theoretical understanding of how gender equality progresses within and across families now and in the future.