Programme de la Journée MDSC 21 Juin 2017

• 9h15 -9h45

  Accueil

  Salle des Actes Grand Château Valrose

• 9h45 - 10h15

  Modelling and Formal Verification of Neuronal Archetypes Coupling.

  (Elisabetta De Maria)

  There exists many ways to connect two, three or more neurons together to form different graphs. We call archetypes only the graphs whose properties can be associated with specific classes of biologically relevant structures and behaviors. These archetypes are supposed to be the basis of typical instances of neuronal information processing. To model different representative archetypes and express their temporal properties, we use a synchronous programming language dedicated to reactive systems (Lustre). The properties are automatically validated thanks to a model checker supporting data types (kind2). The language Lustre and the model checker kind2 are then exploited to investigate the behaviour of the composition of the presented archetypes.

  Slides

• 10h15 - 10h45

  From structural data to in vivo toxicity prediction: a challenge for machine learning.

  (Ingrid Grenet)

  Computational tools are one of the alternative solutions envisioned to assess chemical toxicity potential as early as possible and at lower animal and money cost. One objective is to develop models able to predict in vivo outcomes using existing data from toxicological studies (in vitro, in vivo) and chemical structures. In order to reach such a challenge, we propose to take advantage of both structural and in vitro data to predict in vivo outcomes. Therefore, we describe a global two stages approach to predict in vivo outcomes from molecular structures. The first stage aims at selecting a subset of in vitro assays statistically associated with an in vivo outcome. Next, these assays become the descriptors of a machine learning model predicting the outcome. In the second stage, QSAR models are developed to predict each in vitro assay of the previous stage, based on molecular structures. Finally, we chain up these two types of models to predict the in vivo outcome for a new molecule: First we use the QSAR models to predict in vitro bioactivities based on the new molecular structure. Secondly, these predictions are used as input of the in vivo outcome predictive model.

  Slides

• 10h45 - 11h00

  Pause Café.

• 11h00 - 11h30

  A cat algorithm for 2-polyominoes.

  (Enrico Formenti)

  Slides

• 11h30 - 12h00

  A synergic approach to the minimal uncompletable words problem.

  (Julien Provillard)

  A finite language X over an alphabet is complete if any word is a factor of a word in X. A word which is not a factor of X is said uncompletable. Among them, some are minimal as all their proper factors belong to Fact(X). The problem is to find bounds on the length of the shortest minimal uncompletable words depending on k, the maximal length of words in X. Though Restivo’s conjecture stating an upper bound in 2k^2 was already contradicted twice, the problem of the existence of a quadratic upper bound is still open. Our approach is original and synergic. We start by characterizing minimal uncompletable words. An efficient algorithm is given to speed up the search of such words. Finally, a genetic algorithm using a SAT-solver allows us to obtain new results for the first values of k.

  Slides

• 12h00 - 13h00

  Pause Déjeuner.

  Salle à manger Grand Château Valrose

• 13h15 - 14h15

  Quand l'espace gagne du temps.

  (Jean Charles Regin)

  Cet exposé aborde la question suivante: comment peut-on utiliser la mémoire disponible afin d'accélérer la résolution de problèmes d'optimisation combinatoire. Nous commencerons par rappeler les principes des méthodes existantes qui proposent de sauver des calculs réalisés en mémoire afin d 'éviter de les répéter, comme la programmation dynamique ou qui proposent de travailler directement sur des grandes structures de données, comme les diagrammes de décisions. Après une brève étude des travaux majeurs réalisés dans le cadre des MDD (multi-valued decision diagram) , nous donnerons une idée des nouveaux algorithmes que nous avons développés. Nous montrerons alors que grâce à eux nous pouvons entrevoir une nouvelle façon de combiner l'utilisation mémoire et les calculs qui ne se focalise plus seulement sur un des deux aspects mais introduit un compromis d'utilisation plus réaliste. Nous présenterons alors différents nouveaux principes de modélisation. L'efficacité de notre approche sera montrée sur deux applications réelles. Ensuite, nous donnerons des perspectives d'évolution de notre approche. Enfin, nous conclurons.

  Slides

• 16h00 - 17h00

  Réunion d'équipe.