INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
JoLA: Job Landscape Aware Job Recommendation
No full textInternational audienceJob recommendation (JR), among the most critical challenges of AI, aims to alleviate frictional unemployment with major potential impacts on society and economy at large. However, Job Recommender Systems (JRS) might become counter-productive and create a congestion phenomenon, if job seekers are mostly recommended the most popular job ads.This paper proposes a novel perspective on JRS, observing that the job market tends to involve a number of so-called "orphan" job ads, that receive very few or no applications. The orphan-job phenomenon is detrimental to the job market as it mechanically decreases the number of jobs effectively considered, worsening the market imbalance and increasing the congestion; in the long term, it also tends to prevent companies from publishing other ads, de facto creating a sleeping job market that is not revealed to the job seekers.This paper introduces new JRS losses, aimed to prevent both the congestion and the orphan-jobs phenomenon, based on a differentiable approximation of the market share attributed to a job ad. The resulting so-called Job Landscape Aware recommender system (JoLA) is experimentally assessed and compared with the state of the art on public datasets, showing new trade-offs that exist between standard recommendation metrics and congestion, while enforcing the desired exposure for most ads. The JoLA code is publicly available at https://codeberg.org/solal/jola.</p
PiNodes in the Druid Meta-Compiler
No full textInternational audienceConditional branches incur high runtime overheads: they disrupt the processor's pipeline and add complexity to the control flow, preventing optimizations. Moreover, after inlining transformations and runtime-inserted type checks, it is often the case that some of these branches are redundant. Thus, there are opportunities to eliminate such branches without changing the semantics of the program.In this paper, we explore methods of eliminating such branches in the Druid meta-compiler, a project intended for source-to-source ahead-of-time generation of baseline JIT compilers from a language interpreter. Moreover, we compare our new solution against a pre-existing algorithm in Druid. We extend its intermediate representation with PiNodes, a sparse representation of constraints on SSA variables. Such constraint information is leveraged by the compiler to remove redundant branches. We propose two PiNode-based methods of flow-sensitive analysis: one based on modelling possible constant values for variables, and a re-implementation of the traditional ABCD algorithm
Efficient data assimilation of unsteady 3D turbulent flows using intrusive generative models
No full textInternational audienc
Cryptanalyse de schémas de cryptographie à clé publique
No full textCryptanalysis of public-key cryptosystems is based on algorithmic and algebraic techniques in number theory. In the first part of this thesis, we present improvements to the LLL algorithm, originally designed by Lenstra, Lenstra, and Lovasz in the eighties to reduce a Euclidean lattice, i.e. to reduce the norm and orthogonalize the basis vectors as possible. We also show how to use this algorithm to reduce rank-2 module lattices in a cyclotomic number field with subfields using symplectic matrices. Indeed, some schemes such as NTRU or Falcon, whose security relies on this difficult problem, have been proposed in post-quantum cryptography and for homomorphic encryption. We also improve linear algebra techniques and propose better algorithms when the matrix is diagonally dominant. These advanced methods allow us to set new records in the computation of number fields: class number, generators of the unit group, generator of a principal ideal. In the second part, we study various classical problems in number theory: we improve algorithms for testing the primality of an integer, and in particular, the cyclotomic test initially proposed by Adleman and developed by Mihailescu. Then, we study different algorithms in a so-called black-box ring model, i.e. we study the number of additions and multiplications in the ring, without looking at how the elements are represented and how the computations are performed in the ring. This allows us in the final chapter, to instantiate these algorithms in different rings in order to propose efficient algorithms for cryptanalysis. In doing so, we are able to more easily distribute the computations of the entire algorithm, while so-called index-calculus algorithms use a linear algebra step which is difficult to parallelize.La cryptanalyse de schémas de cryptographie à clé publique repose sur un ensemble de techniques algorithmiques et algébriques en théorie des nombres. Dans une première partie de cette thèse, nous présentons des améliorations de l’algorithme LLL, dû à Lenstra, Lenstra et Lovasz pour réduire un réseau euclidien, c’est-à-dire réduire la norme et orthogonaliser le plus possible les vecteurs de la base. Nous montrons aussi comment utiliser cet algorithme pour réduire des réseaux modules en rang 2 dans un corps de nombres cyclotomique ayant des sous-corps. En effet, certains schémas comme NTRU ou Falcon, dont la sécurité repose sur ce problème difficile, ont été proposés en cryptographie post-quantique et pour du chiffrement homomorphe. Nous améliorons aussi les techniques d’algèbre linéaire creuse et proposons de meilleurs algorithmes lorsque la matrice est à diagonale dominante. Ces avancées nous permettent de réaliser de nouveaux records de calculs de corps de nombres : nombre de classes, générateurs du groupe des unités, générateur d’un idéal principal. Dans une seconde partie, nous étudions différents problèmes classiques en théorie des nombres : nous améliorons différents algorithmes pour tester la primalité d’un entier et en particulier, le test cyclotomique initialement proposé par Adleman et dernièrement développé par Mihailescu. Puis, nous étudions différents algorithmes dans un modèle dit de l’anneau en boîte noire, c’est-à-dire que nous étudions le nombre d’additions et de multiplications dans l’anneau, sans nous intéresser à la façon de représenter et de faire les calculs dans cet anneau. Ceci nous permet dans le dernier chapitre, d’instancier ces algorithmes en fonction de différents anneaux pour proposer des algorithmes efficaces en cryptanalyse. Ce faisant, nous sommes capables de distribuer plus facilement les calculs de tout l’algorithme, alors que les algorithmes dit de calcul d’indice utilisent une étape d’algèbre linéaire qu’il est difficile de paralléliser
Détection et Reconstruction de Nuages de Points Plénoptiques
No full textThis thesis provides an overview of the plenoptic function and how it relates to volumetric data, through plenoptic scene representations. It reviews the existing methods that introduce such capability explicitly and implicitly, in the form of a Plenoptic Point Cloud (PPC) and Radiance Fields (RF), respectively. These are presented from the point of view of the challenges of practicality of these methods for applications of streaming for interactive content, namely the size and speed. Initially, we incorporated the plenoptic capability for MPEG’s geometry-based encoder (G-PCC) by compressing PPCs using a combination of the linear transforms over the color vector of the different camera viewpoints combined G-PCC’s predictive attribute coders. We follow this by addressing the size disadvantages of real-time rendering implementations of NeRF-based methods, by introducing a compression pipeline to the PlenOctrees model. Afterwards, we introduce a method to systematically generate PPCs and compare them directly against RF solutions with conventionally used render-based metrics. Finally, we leverage the underlying geometry of the RF models to orient their pruning for a more efficient compression.Cette thèse donne un aperçu de la fonction plénoptique et de la manière dont elle est liée aux contenus volumétriques, par le biais de représentations de scènes plénoptiques. Elle passe en revue les méthodes existantes qui introduisent cette capacité de manière explicite et implicite, sous la forme d’un nuage de points plénoptique (PPC) et de champs de radiance (RF), respectivement. Ces méthodes sont présentées du point de vue des défis qu’elles posent en termes de praticité pour les applications de diffusion en continu de contenu interactif, à savoir la taille et la vitesse. Dans un premier temps, nous avons intégré la capacité plénoptique pour le codeur MPEG basé sur la géométrie (G-PCC) en compressant les PPC à l’aide d’une combinaison des transformées linéaires sur le vecteur de couleur des différents points de vue de la caméra combinées aux codeurs d’attributs prédictifs du G-PCC. Nous abordons ensuite les inconvénients liés à la taille des implémentations de rendu en temps réel des méthodes basées sur le NeRF, en introduisant un pipeline de compression dans le modèle PlenOctrees. Ensuite, nous introduisons une méthode pour générer systématiquement des PPC et les comparer directement aux solutions RF avec des mesures conventionnelles basées sur le rendu. Enfin, nous tirons parti de la géométrie sous-jacente des modèles RF pour orienter leur élagage en vue d’une compression plus efficace
Implementation and short-term outcomes of minimally invasive liver surgery in France
No full textInternational audienc
RDF Query Answering in the Presence of Access Restrictions
No full textIn this work, we explore algorithms for answering conjunctive RDF queries in the presence of RDFS ontologies and access control. We consider an access control setting where by default all users have access to the complete graph, and a restriction can forbid user a user's access to specific IRIs. Here, restricting for user u the access to an IRI i entails that: no answer to a query by u may contain the IRI i; no triple containing i can be used to compute an answer for a query by i, nor to entail such a triple via reasoning with the ontology. We present a set of query answering algorithms for this novel context, and prove that five among them are correct, i.e., sound and complete, with respect to both the ontology and the access restrictions in place. We have implemented all our algorithms and present experiments comparing their performance
Real-time Photorealistic Mapping for Situational Awareness in Robot Teleoperation
No full textInternational audienceAchieving efficient remote teleoperation is particularly challenging in unknown environments, as the teleoperator must rapidly build an understanding of the site's layout. Online 3D mapping is a proven strategy to tackle this challenge, as it enables the teleoperator to progressively explore the site from multiple perspectives. However, traditional online map-based teleoperation systems struggle to generate visually accurate 3D maps in real-time due to the high computational cost involved, leading to poor teleoperation performances. In this work, we propose a solution to improve teleoperation efficiency in unknown environments. Our approach proposes a novel, modular and efficient GPU-based integration between recent advancement in gaussian splatting SLAM and existing online map-based teleoperation systems. We compare the proposed solution against state-of-the-art teleoperation systems and validate its performances through real-world experiments using an aerial vehicle. The results show significant improvements in decision-making speed and more accurate interaction with the environment, leading to greater teleoperation efficiency. In doing so, our system enhances remote teleoperation by seamlessly integrating photorealistic mapping generation with real-time performances, enabling effective teleoperation in unfamiliar environments.</div
Investigating Language Model Capabilities to Represent and Process Formal Knowledge: A Preliminary Study to Assist Ontology Engineering
No full textInternational audienceRecent advances in Language Models (LMs) have failed to mask their shortcomings particularly in the domain of reasoning. This limitation impacts several tasks, most notably those involving ontology engineering. As part of a PhD research, we investigate the consequences of incorporating formal methods on the performance of Small Language Models (SLMs) on reasoning tasks. Specifically, we aim to orient our work toward using SLMs to bootstrap ontology construction and set up a series of preliminary experiments to determine the impact of expressing logical problems with different grammars on the performance of SLMs on a predefined reasoning task. Our findings show that it is possible to substitute Natural Language (NL) with a more compact logical language while maintaining a strong performance on reasoning tasks and hope to use these results to further refine the role of SLMs in ontology engineering.</div
ECSPLAIN: Explainability Constrained-claSsifier for Pairing the detection and the Localization of moving Areas from SAR INterferograms
No full textInternational audienceDetecting slope instabilities on synthetic aperture radar (SAR) interferograms using deep learning approaches presents several challenges. This detection task suffers from the lack of transparency of deep networks, the complexity of the input data (i.e., complex values, sensitivity to distortions, and presence of counterfactuals), and the complexity of the target phenomena (i.e., the variable velocities and the complex underground processes). In this article, we propose a new framework called explainability-constrained classifier for pairing the detection and the localization of moving areas on interferograms (ECSPLAIN), to generate decision, localization, and segmentation maps from a single but explainable classifier network. It consists of training a classifier to detect whether an instability is located in the patch or not, and to explain its decision with a class activation map (CAM) that matches the actual location of the instability. Therefore, by using a single classifier network, the framework can pair the detection and the localization of moving areas. Four CAMs are investigated for the training of the ECSPLAIN framework. Experiments on the ISSLIDE dataset show that our proposal achieves better explainability than standard a posteriori CAMs with more than 0.20 points of improvement in terms of Dice and IoU scores. It also allows competitive performance with segmentation-only networks, with only 0.04 points of difference in terms of Dice and intersection over union (IoU) scores. Thus, the proposed method is competitive with the most efficient methods while being lighter, faster, and delivering a decision based on a human-like reasoning process. Finally, the ECSPLAIN framework is applied to enrich the ISSLIDE dataset, discovering more than 470 manually validated slope instabilities over the Alps