Plenary Speakers
- Nozha Boujemaa (Inria Saclay Ile-de-France)
- Christophe Cérin (Université Paris 13)
- Sylvie Méléard (Ecole Polytechnique, France)
- Philippe de Reffye (CIRAD, France)
- Hatem Zaag (Université Paris 13)
Nozha Boujemaa
Toward Responsible and Ethical Smart Data Research and Innovation.
Abstract: The advent of massive data is pushing the limits in many areas. Mainstream statistical models and decision-making algorithms are challenged by heterogeneous, semi-structured, complex, incomplete and/or uncertain data. In the last few years, Deep Learning achieved quite a few breakthroughs in Data Science, training deep neural nets from complex raw data (image, signal, natural language, control). Cognitive approaches arise enhancing computer-aided decision making toward "Artificial Intelligence" era. On the other hand, data carry with it new operability constraints in terms of privacy, security and integrity -- depending on the application domains. The digital world tends to establish an asymmetry of information between citizens and public authorities on one hand and private companies on the other hand with respect to collection and processing of personal data. This asymmetry creates a mistrust: fueled by hidden data usages, dissemination practices escaping the control of individuals, business models based on data over-collection – the whole framed by an obsolete regulation. Economic, political and social consensus is emerging to build trust for resolving the dilemma between the expected benefit of innovations exploiting these new potentials and the threats on privacy. Nowadays, the economic success of digital technologies on markets depends on consumers’ appropriation, and trust is a key concept explaining this process – including its first step of adoption. In some sensitive areas, it is important to know where the data is coming from and how it was processed in the big data pipeline. We will develop three challenging research dimensions where "Trust & Transparency" methods and tools should be developed to foster trustworthiness in big data technologies.
Christophe Cérin
Abstract: In this presentation we would like to make a synthesis of researches on cloud and grid computing that we started near 10 years ago, partly with Tunisian colleagues. We explain first the scientific challenges and issues for leveraging the technology, then we give details about our propositions and contributions related to BonjourGrid, PastryGrid, RedisDG, SlapOS, Wendelin middleware. The two last projects are industrial projects… and I will explain some 'ways of thinking' in this eco-system related to Open Source. The first common thread of our work is to decentralize as much as possible the computing infrastructure by a decoupling of the main components in interaction. The second one is to investigate the coupling of a scientific problem and a testbed under the umbrella of the scientific experimental approach. The ‘scientific experimental approach’ is when we investigate the whole chain, starting from the modeling, the analysis, the design and the implementation at large scale, the return from experiments to formulate new hypotheses. Many experimental results for the validation of our work have been conducted on the Grid5000 testbed and are summarized on http://lipn.univ-paris13.fr/bigdata. Information under this link surpasses cloud problems to open to issues on Systems for Big Data.
Sylvie Méléard
Stochastic dynamics in adaptive biology
Abstract: Understanding the adaptation and evolution of populations is a huge challenge, in particular for microorganisms since it plays a main role in the virulence evolution or in bacterial antibiotics resistances. We propose a general eco-evolutionary stochastic model of population dynamics with clonal reproduction and mutation. Moreover the individuals compete for resources and exchange genes, as in the transfer of plasmids in bacteria. We study different asymptotics of this general birth and death process depending on the respective demographic, ecological and transfer time-scales and on the population size. We explain how the gene transfer can drastically affect the evolutionary outcomes. This work is developed with S. Billiard, P. Collet, R. Ferrière and C.V. Tran.
Philippe de Reffye
Data assimilation and parameter estimation on plant growth models with stochastic development.
Background and aims: FSPM implementations (platforms) are mainly simulation oriented, allowing hypothesis testing and insights, but lacking data assimilation and inverse methods (parameter identification and estimation) capabilities, especially when stochastic development is considered. However, in the past decade, major progress were hold on both structural parameter retrieval (using for instance crown analysis) and functional parameter estimations, but limited to single monoculm plants or plants showing simple deterministic structures. We propose hereby a model and methodological frame allowing efficient data assimilation and FSPM parameter estimation on stochastic ramified plants.
Methods: Most of plants show stochastic traits in their architecture, resulting from the buds activity during the development process, related to breaks, abortions, or ramification delays. We model plant architecture as a set a ramified axis, composed of phytomers series with their subsets of organs (leaves, internode …). Stochastic plants construct from so called stochastic development axis in which each phytomer is given a probability of break, a probability of reliability and a probability of branching. We introduce the concept of potential structure, as the structure which contains all possible plant realizations in which phytomers are weight by their probability to appear. On the functioning point of view, organs sharing the same physiological age and ontologic age, build cohorts sharing also the same fate. An organic series is the set of organs of same nature produced by a same apical meristem along an axis (bellowing to the same stochastic development axis). Organic series are orthogonal to organ cohorts. The data assimilation process builds organic series issued from a set of stochastic simulations. These series are fitted with the corresponding series computed from the potential structure, using parameter identification provided by the generalized nonlinear least square method parameter estimation.
Results and Conclusions: The concept of potential structure, defining the mathematical domain of a stochastic computational plant structure builds a novel conceptual frame. On a limited set of simulated plants, from few organic series the inverse method retrieved the parameter values corresponding to the origin set of parameters used for the simulations. The data assimilation operated by the organic series extraction is promising, since this in silico process repeats exactly the procedure to apply on real plants selected from a stand. Applications on real plants are in progress showing first interesting results on source and sink parameter estimation. Requiring the concept of physiological age and the assumption of the common biomass pool, the proposed approach is versatile; it doesn’t rely on the plant architectural model nor its structural stochastic expression. Identifying parameters on complex plant architectures opens a new range of coming applications such as phenotyping and crop routing optimisation.
Keywords: GreenLab, plant architecture, source-sink parameters, model calibration, stochastic process, data-assimilation, inverse model, functional-structural plant model.
Hatem Zaag
Titre : Solitons et singularités pour l'équation semi-linéaire des ondes.
Résumé : Sous certaines conditions, l'équation semilinéaire des ondes avec nonlinéarité sous conforme donne lieu à une solution qui n'existe pas pour tout temps. On dit alors qu'elle explose en temps fini. Par la vitesse de propagation finie, la solution explose partout en espace, mais à des temps différents. Deux types de d'explosion se dégagent alors, selon que l'on soit au voisinage d'un point "non-caractéristique" ou d'un point "caractéristique". Dans cas non-caractéristique, moins singulier, on peut définir une notion de profil à l'explosion, qui est en fait un "soliton" lié à celui de l'équation de Korteweg-de-Vries. Dans le cas "caractéristique", plus singulier, on observe au moins 2 solitons qui se repoussent. Ces solitons sont évidemment "alignés" en dimension 1, mais en dimension 2, ils peuvent être disposés en croix, donnant lieu à un ensemble d'explosion en forme de pyramide.