Projects

Experimaestro is an experiment manager based on a server that contains a job scheduler (job dependencies, locking mechanisms) and a framework to describe the experiments with JavaScript or in Java.

A common way to achieve scalability for processing SPARQL queries over large RDF data sets is to choose map-reduce frameworks like Hadoop or Spark. Processing complex SPARQL queries generating large join plans over distributed data partitions is a major challenge in these shared nothing architectures. In this article we are particularly interested in two representative distributed join algorithms, partitioned join and broadcast join, which are deployed in map-reduce frameworks for the evaluation of complex distributed graph pattern join plans. We compare five SPARQL graph pattern evaluation implementations on top of Apache Spark to illustrate the importance of cautiously choosing the physical data storage layer and of the possibility to use both join algorithms to take account of the existing predefined data partitionings. Our experimentations with different SPARQL benchmarks over real-world and synthetic workloads emphasize that hybrid join plans introduce more flexibility and often can achieve better performance than join plans using a single kind of join implementation.

Block-o-Matic is a web page segmentation algorithm based on an hybrid approach for scanned document segmentation and visual-based content segmentation. A web page is associated with three structures: the DOM tree, the content structure and the logical structure. The DOM tree represents the HTML elements of a page, the geometric structure organizes the content based on a category and its geometry and finally the logical structure is the result of mapping content structure on the basis of the human-perceptible meaning that conforms the blocks. The segmentation process is divided in three phases: analysis, understanding and reconstruction of a web page. An evaluation method is proposed in order to perform the evaluation of web page segmentations based on a ground truth of 400 pages classified into 16 categories. A set of metrics are presented based on geometric properties of blocks. Satisfactory results are reached when comparing to other algorithms following the same approach.

Coriolis is an experimental integrated platform for the research, development and evaluation of new back-end VLSI design flows. Interconnect scaling to nanometer processes presents many difficult challenges to CAD flows. Currently academic research on back-end tend to address only specific algorithmic issues separately, although one key issue to address is the cooperation of multiple algorithmic tools. CORIOLIS, our platform, is based on an integrated C++ database around which all tools consistently interact and collaborate. This platform currently includes a timing-driven global place and route flow.

Our purpose is to provide a language for designing generators of analog functions, that can be easily ported to new set of specfications and new technologogy processes. We are currently developing such a language that is called CAIRO+ The CAIRO+ language supports the four steps of a design flow based on net-list and layout templates. This language is aimed to help the designer to capture his knowledge, thus creating a library of layout-aware analog functions. It is based on C++ language. The design flow relevant to CAIRO+ is the following : ->net-list and layout template capture, ->design space exploration (managing electrical constraints) ->shape function computation (managing geometrical constraints) ->layout generation (place and route) CAIRO+ allows creating complex hierarchical analog function generators by using existing generators of simpler functions. It is an answer to the problem of Analog and Mixed IPs. As a demonstration of the CAIRO+'s capabilities, we are developping Analog to Digital converters, specially Sigma Delta.

Lemon is a GUI toolkit written in C++ for manipulating graphical models. It relies on aGrUM both for graph theoretic algorithms and for computations within the graphical models (e.g., inference, learning).

aGrUM is a C++ library designed for easily manipulating graphical models. Its range of applications is quite large as it is designed, e.g., for performing learning tasks (for instance, learning Bayes nets from data), planning tasks (FMDPs) and inference (Bayes nets, GAI-nets, influence diagrams).

In the general context of the field of eXplainable Artificial Intelligence (XAI), the IFP-in-RL project aims to propose a method for the automatic construction of a control system of a system, such as a drone, which takes take into account the interpretability constraint in its very design. This project takes place within the framework of systems based on fuzzy rules which, since their introduction, aim to facilitate the expression of knowledge in a linguistic form, natural for the user, and easily understandable by a human. Such a knowledge representation is an excellent way to promote human interaction with the computer system and to improve their understanding of how it works, thus offering the possibility of making their behavior transparent and easily validated. In the literature, different approaches to build or to fine-tune a fuzzy rule base to design a system exist, but they generally suffer from the drawback of not incorporating specific interpretability optimization. In this project, an innovative methodology is introduced for the design of such systems. This methodology is based on the implementation of a reinforcement learning approach using interpretability metrics. The objective here is to integrate the consideration and optimization of the desired interpretability during the learning itself, and not a posteriori as many methods currently do in the field of XAI. The IFP-in-RL project aims to achieve this upstream, a complete study, both theoretical and experimental, of interpretability metrics, including existing numerical criteria as well as user needs. This will involve proposing a taxonomy of existing metrics and defining new measures if necessary, in order to complete the previous ones and allow their exploitation in original reinforcement learning algorithms. An original feature of this project is to integrate a qualitative assessment, carried out on a human panel, of the proposed metrics but also of the rule bases obtained at the end of reinforcement learning. In application terms, the objective of the IFP-in-RL project is to implement these proposals for piloting a drone, navigating in complete autonomy to ensure a mission consisting of flying over points of interest and taking pictures, from data provided by a simulator.

obtenu dans le cadre du Partenariat d'Innovation Intelligence Artificielle (P2IA) du ministère de l'éducation nationale et porté par la startup EvidenceB, implique des entreprises (Nathan, Daesign, Schoolab, Isograd, BlueFrog), deux laboratoires (LIP6 et Inria Bordeaux), l'APMEP (association des professeurs de mathématiques) ainsi que des chercheurs en psychologie cognitive (E. Sander) et en neurosciences (A. Knopf). Il vise à réaliser un assistant pédagogique pour les mathématiques du Cycle 2 (CP, CE1, CE2) s'appuyant sur des algorithmes d'IA et sur un ensemble d'exercices définis à partir d'avancées en sciences cognitives. Nous travaillons sur une brique IA visant à proposer des regroupements d'élèves (textit{clustering}) appris sur l'ensemble des classes sur la base de critères de maîtrise de compétences en mathématiques. Ce textit{clustering} est ensuite appliqué classe par classe à intervalles réguliers pour proposer à l'enseignant un suivi de l'évolution de ses groupes d'élèves, afin de faciliter la mise en place de stratégies de pédagogie différenciée.

MindMath (financé par la BPI et la région IdF) est un projet de plateforme gamifiée et adaptative pour l’apprentissage des mathématiques au collège. Ce projet implique plusieurs partenaires industriels (Cabrilog, Tralalere, Domoscio et Bayard) et académiques (LDAR - Université Paris Diderot et LIP6 - Sorbonne Université). Nous développons des algorithmes pour décider, en fonction des activités des élèves au sein de tâches de résolution de problèmes en mathématiques, du feedback le plus adapté pour les aider à progresser dans leurs apprentissages. La décision s’appuie à la fois sur une ontologie construite avec des experts en didactique des mathématiques et sur des approches d’apprentissage automatique. La recherche des feedbacks optimaux se fait par apprentissage par renforcement, avec un système de récompense basé sur la réussite des élèves dans les activités. Ces propositions sont expérimentées dans différents contextes scolaires et parascolaires.

IECARE est un projet de recherche financé par l'ANR. Il vise à produire des connaissances fondamentales et opératoires sur l’informatique, son enseignement et son apprentissage, à l’école obligatoire. Ce projet pluridisciplinaire associe des chercheurs en Informatique et en sciences humaines et sociales (Sciences de l’éducation, didactiques, psychologie des apprentissages, sociologie). La recherche suit trois thèmes structurant : analyser les représentations et les pratiques des enseignants et des élèves ; modéliser, concevoir et modifier des scénarios pédagogiques et des ressources pour soutenir les pratiques d’enseignement et d’apprentissage ; étudier les cadres d’accompagnement mis en place par et pour les enseignants et les formateurs en informatique.

MAGAM is a Multi-Aspect (didactic, pedagogic, affective and motivational, gaming, etc.) Generic Adaptation Model based on matrix calculation that aims to adapt learning activities.

This project aims to built a plateform to assist teachers in adapting learning activities. Multiple indicators (cognitive, pedagogical, temporal, etc. will be inferred from data traces that are recorded and generated automatically or manually from the learner activities. Visualization systems will be proposed to assist teachers in their activities' adaptation process and make it.

The objective of the Hubble ANR project is to create a national observatory for the building and sharing of massive data analysis processes, from traces left in e-learning environments. Hubble will enable to analyze and explain phenomena of teaching and learning with these environments.

RecoMOOC aims at reducing attrition and improving the learning experience for MOOC learners through interaction with other students they can talk to and work with, identified automatically through a peer recommender system analyzing each learner's individual factors and progression.

This research project and its associated software aim at designing and implementing a software component model for cyber-physical control systems. It develops over the BCM4Java project from which it uses the basic concepts and the implementation of distributed components in Java (10.000 lines of code and documentation today). Besides integrating real time components, this project studies component-based software architectures for control, their specification using stochastic hybrid systems and their simulation using models following the DEVS standard. A particular focus is given on the parallel composability between components, their individual specifications and simulation models. A software subproject of BCMCyPhy proposes a new implementation in Java of the DEVS standard for modular simulation of components and their assemblies (20.000 lines of code and documentation today). The simulators obtained through the composition of the components simulator models allow to debug, test, verify and validate applications. This software has been used by a few tens of students and is still being used in the context of the ALASCA master 2 course since 2018.

PNML Framework is a prototype implementation of ISO/IEC-15909 part 2, International Standard on Petri Net Markup Language. The primary purpose of PNML is to enable interoperability among Petri net tools. PNML framework has thus been designed to back the Standard. It will enable Petri nets tools developers to seamlessly integrate PNML support into their tools. It provides an extensive and comprehensible API to create, save, load and browse PNML models.

is a Petri Net based CASE environment. It offers a set of services to perform specification, validation, formal verification, model checking, compute structural properties (invariants, traps, syphons etc.) simulate and generate code. These services have been implemented either by members of our team or university partners (Technical university of Helsinki, University of Torino, Technical university of Munchen, Bell laboratories). The second geration of CPN-AMI, build on top of FrameKit, is available on the Internet since March 1997.

SPOT (Spot Produces Our Traces) est une bibliothèque de model-checking facilement extensible. À la différence des model-checkers existants, dont le mode opératoire est immuable, SPOT fournit des briques que l'utilisateur peut combiner entre elles pour réaliser un model-checker répondant à ses propres besoins. Une telle modularité permet d'expérimenter facilement différentes combinaisons, et facilite le développement de nouveaux algorithmes. D'autre part, cette bibliothèque est centrée autour d'un type d'automates particulier permettant d'exprimer les propriétés à vérifier de façon plus compacte, qui n'a jamais été utilisé dans un outil jusqu'à présent.

Remote interoperability testing services for IoT devices

HPDDM is a collection of preconditioners based on domain decomposition, either overlapping or non-overlapping. They can be used to solve large linear systems, as typically encountered when discretizing partial differential equations. These preconditioners can be used in conjunction with various Krylov methods. The library is usable in C, C++, Python, or Fortran codes.

FiXiF is a suite of tools used to implement filters on embedded devices (usually DSP, micro-controllers, FPGA or ASIC) with finite-precision impact in minds (fixed- or floating-point arithmetic).

PROMISE is a tool to auto-tune the precision of floating-point variables in numerical codes.

ExBLAS aims at providing algorithms and implementations for fundamental linear algebra operations (like those included in the BLAS library) that deliver reproducible and accurate results with small or without losses to their performance on modern parallel architectures.

The SAM library enables rounding error estimation in arbitrary precision programs.

CADNA is a library which allows to perform scientific computations with the estimation and the control of the round-off error propagation.

AppQInfo will provide a world class training in photonic Quantum Information Processing (pQIP), and prepare an excellent cohort of Early Stage Researchers (ESRs) to become the future R&D staff of Europe’s emerging markets in this area. Quantum Information Processing (QIP) is a key ingredient in Europe’s future Quantum Communication Infrastructure; it underpins quantum communications and quantum simulations, the first two pillars of the H2020 Quantum Flagship. QIP will revolutionise information technology, providing higher quality, speed and unconditional security, not possible with classical technologies. AppQInfo focusses on QIP in state-of-the-art integrated photonics, one of five Key Enabling Technologies for European Industry. Integrated photonics permits the creation, manipulation and readout of photonic quantum states in a highly controlled manner, with high speeds and low losses. The broad objectives of AppQInfo are: to create an excellent training of ESRs in the field of pQIP that is both interdisciplinary and intersectoral; to develop innovative, entrepreneurial ESRs with great career prospects; to maximise the exploitation and dissemination of our research; to engage the public through several outreach activities; to consolidate a wide expertise in the field of pQIP; to create a long-lasting collaboration network of top-class research units and industrial entities. Together, our 15 interdisciplinary research projects will work towards feasible long- distance quantum communications from urban-scale networks to satellite-based systems using various data encoding; study quantum photonic circuits towards their quantum transport properties and quantum transforms they implement; exploit these platforms for machine learning applications, such as building all-optical artificial neural networks, and applying them for quantum simulations; develop enabling technologies of sources and detectors of multiphoton quantum states and polaritonic logic gates.

Randomness is a resource that enables applications such as efficient probabilistic algorithms, numerical integration, simulation, and optimization. In the last few years it was realized that quantum devices can generate probability distributions that are inaccessible with classical means. Hybrid Quantum Computational models combine classical processing with these quantum sampling machines to obtain computational advantage in some tasks. Moreover, NISQ (Noisy, Intermediate-Scale Quantum) technology may suffice to obtain this advantage in the near term, long before we can build large-scale, universal quantum computers. PHOQUSING aims to implement PHOtonic Quantum SamplING machines based on large, reconfigurable interferometers with active feedback, and state-of-the-art photon sources based both on quantum dots and parametric down-conversion. We will overview the different architectures enabling the generation of these hard-to-sample distributions using integrated photonics, optimizing the designs and studying the tolerance to errors. We will build two quantum sampling machines with different technologies, as a way to do cross-checks while exploiting all advantages of each platform. These machines will establish a new state-of-the-art in photonic reconfigurability, system complexity, and integration. Finally, we plan to perform first, proof-of-principle demonstrations of Hybrid Quantum Computation applications in optimization, machine learning, and graph theory. The PHOQUSING team includes long-term scientific collaborators who were among the first to demonstrate quantum photonic samplers; two of the leading European start-ups in the relevant quantum technologies; and theoretical experts in photonics and quantum information science. This project will help establish photonics as a leading new quantum computational technology in Europe, addressing the science-to-technology transition towards a new industrial sector with a large foreseeable economic impact.