The mathematical methods and the numerical algorithms developed in the team-projects or in the labs associated with the ISCD are collected and, if needed, optimized by the engineers at ISCD. The software packages are disseminated, mostly as open-source software, into the scientific community. A selection of these packages is given hereafter:
Tinker-HP is a free molecular dynamics software package developed by Prof. Piquemal's group at LCT-ISCD for solving for multiscale simulations of large complex systems with advanced polarizable force fields on parallel computers.
Basilisk is a free software developed by Stephane Popinet et al. at D'Alembert laboratory (SU), for the solution of partial differential equations on adaptive Cartesian meshes.
Mmg is an open source software for simplicial remeshing and mesh adaptation. An open-source consortium project is ensuring that Mmg remains a non-competitive and multidisciplinary tool as well as a research framework.
This toolbox repository provides a collection of open-source finite element solvers for solving Partial Differential Equations on unstructured meshes (fluid mechanics, elasticity, shape optimization, transport).
The team leaders and the researchers at the ISCD have been successful in raising funding through different types of grants, including grants for large collaborative research (Programme Investissements d'Avenir). Over the last five years, the yearly amount of third-party funding represents more than 3 times the endowment of the ISCD.
To further our goal and to foster interdisciplinary research, the Board of Affiliates of the ISCD is pleased to announce a call for proposals "Support to research at the interfaces". Aimed at advancing research that combines computing and/or data science expertise with domain expertise, this funding is intended to
We are particularly drawn to sustain projects that will enable faculty teams to develop successful proposals for large scale grants. We will look for applications that propose unique and novel approaches to bring scholars together to work on projects that cross traditional discipline boundaries.
The affiliates and associated faculty are eligible to request funds to support:
Co-financing (ANR, European contracts, ...) is strongly encouraged. The researchers are invited to study the possibility of co-financing the salaries of doctoral students and post-doctoral students recruited on ISCD credits.
Affiliates and associate members of the institute's team projects are eligible for funding, if it is intended to support actions within the scientific scope of the project, as defined by its scientific leader. These requests have priority over requests from researchers or professors at Sorbonne Université who do not belong to ISCD team projects.
To be announced early 2019.
Funding can be requested for innovative multidisciplinary studies in the field of intensive computing, data analysis, AI, scientific visualization outside the thematic axes of the project teams, in a percentage that will not exceed 20% of the annual budget of the ISCD. For applications outside the team projects, the research-training link will be favored and training actions are strongly encouraged, especially if they rely on collaborative dissemination tools (E-learning, MOOC).
Steps to apply for a financial support are described hereafter (click picture to enlarge):
[wordpress_file_upload uploadpath="funding_upload" fitmode="responsive" uploadpatterns="*.pdf*" maxsize="5" createpath="true" forcefilename="true" uniquepattern="datetimestamp" placements="filename+selectbutton+uploadbutton/subfolders/userdata/progressbar/message" uploadtitle="Archive file" notify="true" notifyrecipients="contact_iscd@upmc.fr" notifysubject="File Upload Notification on iscd.upmc.fr" userdatalabel="Your message|t:checkbox|s:left|r:1|a:0|p:inline|d:toto|l:Choose|f:right"]
The Institute for Computing and Data Sciences (ICSD) and its affiliated laboratories and units organize a large number of seminars, workshops, conferences and other events for discussing latest scientific results as well as identifying upcoming challenges in the field of scientific computing and data analysis. In addition the ISCD regularly hosts specific events to widen the interdisciplinary dialogue.
The ISCD Colloquium is a monthly event which gathers scientists from many disciplines to discuss interdisciplinary issues. Renowned experts are invited to present their latest results and discuss the current challenges of their discipline in connection with scientific computing and data sciences. The colloquium is open to scientists from all disciplines, including medicine and humanities. Post-docs, PhDs and research master students are most welcome to attend the lectures as the latter are also intended to connect young scientist to the research community.
The colloquium will be held on fridays at 3:30pm, at Sorbonne Université, amphi Charpak (access hall 22), campus Pierre et Marie Curie (click on the name or title to know more about the subject).
Subscribe/unsubscribe to the announce mailing list through the web interface or via email so you can be apprised of upcoming lectures.
Résumé : Le transport optimal a été formulé par Gaspard Monge au 18e siècle. Il s'agit d'optimiser le coût de transport depuis un ensemble de sources (par exemple les boulangeries) vers des consommateurs (par exemple les cafés, le matin dans Paris). Ce problème très ancien a connu plusieurs révolutions. Léonid Kantorovitch a expliqué en 1942 comment le reformuler en un problème plus facile à résoudre et étudier : il a obtenu le prix Nobel d’économie pour ses travaux. Dans les années 90, Yann Brenier, un mathématicien Français, a établi un lien entre les travaux de Monge et Kantorovitch, et plus tard Cédric Villani a obtenu la médaille Fields (l'équivalent du prix Nobel) notamment pour ses travaux sur le transport optimal. Et depuis quelques années, des mathématiciens et informaticiens ont développé des techniques numériques révolutionnaires pour appliquer le transport optimal à d'innombrables problèmes concrets tels que le traitement d'images et l’intelligence artificielle. Dans cet exposé je ferai un tour d'horizon de ces différentes révolutions.
Bio : Gabriel Peyré est directeur de recherche au CNRS, au département de mathématiques et applications de l’Ecole Normale Supérieure. Ses travaux sont à l’interface entre les mathématiques appliquées, l’informatique graphique, la vision par ordinateur et les neurosciences. Il a obtenu 2 bourses ERC (starting grant SIGMA-Vision en 2010 et consolidator grant NORIA en 2016) ainsi que le prix Blaise Pascal en 2017. Gabriel Peyré est engagé dans la recherche reproductible et l’éducation au code, en particulier à travers la plateforme www.numerical-tours.com.
Résumé : Les images, qui il y a cinquante ans étaient presque toujours imprimées sur du papier ou sur une pellicule, sont en trente ans devenues digitales, et ont proliféré car avec les caméras digitales connectées, il est immédiat de les capter et de les transmettre. Ces images sont prises par les particuliers, mais aussi par tous les professionnels dans tous les secteurs de la science et de la technologie. Il n’est plus maintenant question que de les analyser automatiquement, car une analyse visuelle et manuelle est devenue impossible. Dans cette conférence je ferai le point sur les progrès en analyse automatique d’images, un sujet qui nous renvoie au mystère plus ancien, et non résolu, de la définition de la perception visuelle. J’analyserai plusieurs exemples de techniques d’analyse automatiques inspirées de la théorie de la perception, tels que l’analyse automatique de la perspective dans une image, la détection d’anomalies et de falsifications, la reconnaissance des empreintes digitales, l’extraction de réseau vasculaire en imagerie médicale, etc.
Bio : Jean-Michel Morel a présenté en 1985 son doctorat d’état sur les équations aux dérivées partielles non-linéaires sous la direction de Haïm Brezis à l’Université Pierre et Marie Curie. Il a été successivement assistant à l’Université de Marseille-Luminy, maître-assistant puis professeur à l’Université Paris-Dauphine et professeur à l’Ecole Normale Supérieure de Cachan. Il a été membre junior de l’institut Universitaire de France de 1993 à 1997 et membre sénior de 2010 à 2015. Il a dirigé ou codirigé 45 thèses de doctorat en mathématiques appliquées au traitement d’images et dirige actuellement une trentaine de chercheurs sur l’analyse mathématique et les algorithmes du traitement d’images. Cette équipe a travaillé sur les questions algorithmiques et mathématiques posées par la conception au CNES de satellites d’observation de la Terre (SPOT5, Pléiades, OTOS). Les algorithmes de débruitage d’images inventés par l’équipe ont été installés dans plus de 500 millions de caméraphones par DxO Labs. JMM a fondé en 2011 Image Processing on Line (www.ipol.im), le premier journal publiant des algorithmes et du logiciel reproductibles et des articles exécutables en ligne. IPOL a des collaborateurs dans 15 universités. Ses archives publiques contiennent 150000 expériences en ligne.
Reconnaissances récentes : 2010 Clay Scholar in Residence, ERC advanced grant 2010, Grand Prix INRIA-Académie des Sciences 2013, Médaille de l’innovation du CNRS 2015, IEEE Longuet-Higgins prize 2015. Les intérêts principaux de l’équipe de recherche de Jean-Michel Morel sont la restauration d’images et de vidéo, la stéréo-vision et la reconstruction du relief de la Terre, la théorie et les applications de la détection de structure et d’évènements dans les images et films, la synthèse d’images et de textures.
Summary: We build branched DNA species that can be joined using Watson-Crick base pairing to produce N-connected objects and lattices. We have used ligation to construct DNA topological targets, such as knots, polyhedral catenanes, Borromean rings and a Solomon's knot.
Nanorobotics is a key area of application. We have made robust 2-state and 3-state sequence-dependent programmable devices and bipedal walkers. We have constructed 2- dimensional DNA arrays with designed patterns from many different motifs. We have used DNA scaffolding to organize active DNA components. We have used pairs of 2-state devices to capture a variety of different DNA targets. We have constructed a molecular assembly line using a DNA origami layer and three 2-state devices, so that there are eight different states represented by their arrangements. We have demonstrated that all eight products can be built from this system. Recently, we connected the nanoscale with the microscale using DNA origami.
We have self-assembled a 3D crystalline array and reported its crystal structure to 4 Å resolution. We can use crystals with two molecules in the crystallographic repeat to control the color of the crystals. Rational design of intermolecular contacts has enabled us to improve crystal resolution to better than 3 Å. We can now do strand displacement in the crystals to change their color, thereby making a 3D-based molecular machine; we can visualize the presence of the machine by X-ray diffraction.
The use of DNA to organize other molecules is central to its utility. Earlier, we made 2D checkerboard arrays of metallic nanoparticles, and have now organized gold particles in 3D. Most recently, we have ordered triplex components and a semiconductor within the same lattice. Thus, structural DNA nanotechnology has fulfilled its initial goal of controlling the internal structure of macroscopic constructs in three dimensions. A new era in nanoscale control awaits us.
Anouk Barberousse | Gérard Biau | Pierre-Yves Boëlle | François Bouchet |
Anne Bourdon | Julien Brajard | Frédérique Charles | Emmanuel Giner |
Katell Guizien | Elodie Laine | Stéphane Popinet | Marco Saitta |
Xavier Tannier | Jean-Pierre Van Staevel | Amélie Viallet |
Le but de ces rencontres informelles est de réunir des gens d'horizons différents, intéressés par les aspects mathématiques de la visualisation.
Pour toute question ou suggestion, contactez l'organisateur Pierre-Antoine Guihéneuf.
À 14h00 en salle 33-34-201
Yacine Bouzidi (INRIA),
A Symbolic Approach for the $H_\infty$ control problem.
Résumé
A classical and thoroughly studied problem in automatic control theory is the $H_\infty$ control of linear systems. Given a linear dynamical system, the objective is to synthesize controllers that achieve stabilization and guarantee some performance criteria according to the $H_\infty$-norm. In this presentation, given a linear dynamical system that depends on a set of physical parameters, we propose a symbolic approach for studying the $H_\infty$ control problem. This approach, based on solving nonlinear matrix equations known as Algebraic Riccati Equations, reduces the problem to the computation of maximum real solutions of parametric algebraic systems. We then, use classical techniques from real algebraic geometry (Gröbner bases, univariate representations and discriminant varieties) to study the algebraic systems that stem from this problem which allows to exhibit some interesting properties and eases the computation of the solutions.
The presented symbolic approach, which is interesting in the context of adaptive control, is illustrated through a classical example, where explicit formulas are obtained for the robust controller and whose robust margin depends only on the parameters of the systems.
This work is a collaboration with G. Rance and Ar. Quadrat (Safran Electronics & Defense), F. Rouillier (Ouragan, Inria Paris) and Al. Quadrat (NON-A, Inria Lille--Nord Europe).
Lundi 28 mai 2018
Jonas Lukasczyk
Topology-Based Visual Analytics for Extreme Scale Scientific Simulations
Résumé
Recent advances in supercomputing have enabled extreme scale scientific simulations that model complex physical processes. Due to the massive size of these simulations, I/O constraints are the limiting factor for scientific discovery as it becomes impossible to store terabytes of simulation data, or to provide interactive data visualizations. To address these issues, we developed a topology- based visual analytics system that first computes compact analysis and visualization products during simulation runtime, and then later composes these products to enable the interactive exploration of simulation states. The entire system is based on the computation of Merge Trees which are used to identify, track, and render individual simulation features. The core interaction device of the system is a Nested Tracking Graph that illustrates the evolution of these features across time and simulation parameters. The system enables scientists to effectively browse through time, examine different simulation parameters, filter features, and query specific analysis and visualization products in real-time.
Vendredi 18 mai 2018 2018
Maxime Soler
Topologically Controlled Lossy Compression
Résumé
This talk presents a new algorithm for the lossy compression of scalar data defined on 2D or 3D regular grids, with topological control. Certain techniques allow users to control the pointwise error induced by the compression. However, in many scenarios it is desirable to control in a similar way the preservation of higher-level notions, such as topological features, in order to provide guarantees on the outcome of post-hoc data analyses. This paper presents the first compression technique for scalar data which supports a strictly controlled loss of topological features. It provides users with specific guarantees both on the preservation of the important features and on the size of the smaller features destroyed during compression. In particular, we present a simple compression strategy based on a topologically adaptive quantization of the range. Our algorithm provides strong guarantees on the bottleneck distance between persistence diagrams of the input and decompressed data, specifically those associated with extrema. A simple extension of our strategy additionally enables a control on the pointwise error. We also show how to combine our approach with state-of-the- art compressors, to further improve the geometrical reconstruction. Extensive experiments, for comparable compression rates, demonstrate the superiority of our algorithm in terms of the preservation of topological features. We show the utility of our approach by illustrating the compatibility between the output of post-hoc topological data analysis pipelines, executed on the input and decompressed data, for simulated or acquired data sets. We also provide a lightweight VTK-based C++ implementation of our approach for reproduction purposes.
Vendredi 6 avril 2018
Antoine Joux
Système de chiffrement basé sur les premiers de Mersenne
Résumé
Avec l'apparition de l'ordinateur quantique, les systèmes de chiffrement basés sur la factorisation et le problème du logarithme discret risquent de perdre leur capacité à assurer la sécurité des données à protéger. En effet, l'algorithme de Shor permettra de rapidement factoriser et calculer des logarithmes discrets dès lors que l'on pourra disposer d'un ordinateur quantique assez gros.
Dans cet exposé, on présentera un système potentiellement capable de résister à un tel ordinateur basé principalement sur la difficulté de reconnaître des entiers de la forme f*(g^(-1)) mod P (où P est un premier de Mersenne et f et g sont deux entiers ayant peu de 1 dans leur décomposition en binaire) d'un entier aléatoire mod P.
Vendredi 16 février 2018
Pierre-Antoine Guihéneuf
Approximation numérique des ensembles de rotation
Résumé
Cette séance a pour but de lancer un petit projet d'approximation numérique de l'ensemble de rotation des homéomorphismes du tore. Voici ce qu'on pourrait trouver dans ce travail :
de la dynamique topologique sur les surfaces ;
du calcul numérique par de l'arithmétique d'intervalles ;
des algorithmes de recherche de cycles/plus court chemin dans les graphes ;
de la recherche numérique de points périodiques et leur certification via des calculs d’indice ;
de l'optimisation de code.
En pratique, dans l'exposé, je donnerai la définition de l’ensemble de rotation, présenterai rapidement l’algorithme actuellement le moins mauvais (arxiv.org/abs/1702.06190) et donnerai quelques pistes d’améliorations (qui sont nombreuses !).
Vendredi 10 novembre 2017
Mathieu Carrière
Sliced Wasserstein Kernel for Persistence Diagrams
Summary
Persistence diagrams (PDs) play a key role in topological data analysis (TDA), in which they are routinely used to describe topological properties of complicated shapes. PDs enjoy strong stability properties and have proven their utility in various learning contexts. They do not, however, live in a space naturally endowed with a Hilbert structure and are usually compared with non-Hilbertian distances, such as the bottleneck distance. To incorporate PDs in a convex learning pipeline, several kernels have been proposed with a strong emphasis on the stability of the resulting RKHS distance w.r.t. perturbations of the PDs. In this article, we use the Sliced Wasserstein approximation of the Wasserstein distance to define a new kernel for PDs, which is not only provably stable but also discriminative (with a bound depending on the number of points in the PDs) w.r.t. the first diagram distance between PDs. We also demonstrate its practicality, by developing an approximation technique to reduce kernel computation time, and show that our proposal compares favorably to existing kernels for PDs on several benchmarks.
Vendredi 20 octobre 2017
Pierre-Vincent Koseleff
Diagrammes de Chebyshev et nœuds à 2 ponts
Résumé
Avec D. Pecker, nous avons montré que tout nœud de ${\mathbf R}^3 \subset {\mathbf S}^3$ est un nœud de Chebyshev, ie admet une représentation polynomiale de la forme $(T_a(t),T_b(t), T_c(t+\phi))$ où $a,b,c$ sont des entiers, $\phi \in \mathbf Q$ et $T_n$ est le polynôme de Chebyshev $x \mapsto \cos( n \arccos x)$.
Avec F. Rouillier et C. Tran, nous avons proposé un algorithme pour identifier les nœuds de Chebyshev à 2 ponts (cas $a=3$, $(a,b)=1$, $c$ fixé et $\phi$ variant).
L'exposé abordera les différents aspects de ce problème et sa généralisation.
Vendredi 21 avril 2017
Antonin Guilloux
Approche expérimentale des variétés de dimension 3
Résumé
Un cocktail d'arguments théoriques, de considérations combinatoires et de calculs effectifs permet de faire des expériences sur des milliers de variétés de dimension 3, notamment avec le logiciel SnapPy et ses extensions. Je présenterai cette approche expérimentale, en essayant d'en montrer les intérêts, limites et difficultés.
Vendredi 3 mars 2017
Julien Tierny et Gregory Ginot
Homologie persistente
Vendredi 10 février 2017
Séance de présentation