Portail HAL EHESS (École des hautes études en sciences sociales)
Not a member yet
103682 research outputs found
Sort by
Towards Transparency: Exploring LLM Trainings Datasets through Visual Topic Modeling and Semantic Frame
No full textInternational audienceLLMs are now responsible for making many decisions on behalf of humans: from answering questions to classifying things, they have become an important part of everyday life. While computation and model architecture have been rapidly expanding in recent years, the efforts towards curating training datasets are still in their beginnings. This underappreciation of training datasets has led LLMs to create biased and low-quality content. In order to solve that issue, we present Bunka, a software that leverages AI and Cognitive Science to improve the refinement of textual datasets. We show how Topic Modeling coupled with 2-dimensional Cartography can increase the transparency of datasets. We then show how the same Topic Modeling techniques can be applied to Preferences datasets to accelerate the fine-tuning process and increase the capacities of the model on different benchmarks. Lastly, we show how using Frame Analysis can give insights into existing biases in the training corpus. Overall, we argue that we need better tools to explore and increase the quality and transparency of LLMs training datasets
Matching Models and Optimal registry for voluntary Organ Donation registries
No full textNational audienceThis paper considers a general decision model of voluntary organ donation registry. The main example is given by the marrow donors registries organized in most of the major countries. A registry is a list of voluntary donors with known type and transplantation requires identical type between donor and receiver. As typing has a relative high cost registry should be organized in an optimal way in order to increase the probability for a patient to find a donor. This paper shows what is the optimal (but not implementable) registry and how filtering mechanism may be used in order to improve actual registry
Introduction. From Third Gender to Trans and Queer People in Muslim Societies
No full textInternational audienceThe figure of the 'third gender' , is not a 'Western phenomenon' but is present in many Muslim societies from the Medinese mukhannath period until today. While there is a long history of significant roles provided to effeminate men, the opposite -the process of recognition of masculinized females -is much rarer. All the effeminate figures share similar characteristics, such as being young and beardless, and practicing music, singing, and dancing, especially in ceremonies, where they become objects of desire for men. Today, cross-dressing is revisited by Arab choreographers and performing artists, and new medical advances provide third gender individuals the possibility to take hormones and/or undergo gender affirmation surgery to become trans
Teacher Compensation and Structural Inequality: Evidence from Centralized Teach
No full textNational audienceThis paper studies how increasing teacher compensation at hard-to-staff schools can reduce inequality in access to qualified teachers. Leveraging an unconditional change in the teacher compensation structure in Peru, we first show causal evidence that increasing salaries at less desirable locations attracts better quality applicants and improves student test scores. We then estimate a model of teacher preferences over local amenities, school characteristics, and wages using geocoded job postings and rich application data from the nationwide centralized teacher assignment system. Our estimated model suggests that the current policy is both ineficient and not large enough to effectively undo the inequality of initial conditions that hard-to-staff schools and their communities face. Counterfactual analyses that incorporate equilibrium sorting effects characterize alternative wage schedules and quantify the cost of reducing structural inequality in the allocation of teacher talent across schools
Le Journal de la Société des Océanistes en accès ouvert : défis et opportunités pour l’association qui le porte
No full texthttps://www.inshs.cnrs.fr/fr/le-journal-de-la-societe-des-oceanistes-en-acces-ouvert-defis-et-opportunites-pour-lassociation-qu
Représentations et conditions d'adoption de véhicules intermédiaires. Une montée en réalité contrariée ?
No full textInternational audienc
Approche interdisciplinaire des unités phraséologiques (UP) dans les langues du mondeLinguistique - TAL & IA - Traduction - Littérature
No full textInternational audienc
Business cycles fluctuations in three-sector intertemporal equilibrium models
No full textInternational audienceThis paper introduces a novel mechanism driving endogenous business cycle fluctuations within a frictionless three-sector intertemporal equilibrium model. We emphasize the critical role of consumer preferences as a primary driver of cyclical dynamics by considering a consumption bundle composed of a pure consumption good and a mixed consumption-investment good that simultaneously serves as both a final consumption good and a capital-accumulating investment good. Endogenous fluctuations naturally arise from sectoral capital intensity differences, an intertemporal consumption trade-off between the two goods, or the interaction of both mechanisms. We offer a detailed characterization of the economy's dynamics, identifying the Hopf bifurcation conditions that trigger persistent cyclical behavior. Additionally, we explore the periodicity of the resulting limit cycles, providing insights into how shifts in preferences and sectoral complementarities can generate self-sustained macroeconomic fluctuations
