Section: New Results

Decision aiding with models and simulation data

Participants : Louis Bonneau de Beaufort, Tassadit Bouadi, Marie-Odile Cordier, Véronique Masson, René Quiniou.

Models can be very useful for decision aiding as they can be used to play different plausible scenarios for generating the data representing future states of the modeled process. However, the volume of simulation data may be huge. Thus, efficient tools must be investigated in order to store the simulation data, to focus on relevant parts of the data and to extract interesting knowledge from these data.

A datawarehouse for simulation data

The Acassya project 8.2.1 aims at providing experts or stakeholders or farmers with a tool to evaluate the impact of agricultural practices on water quality. As the simulations of the deep model TNT2 are time-consuming and generate huge data, we have proposed to store these simulation results in a datawarehouse and to extract relevant information, such as prediction rules, from the stored data. We have devised a general architecture for agro-environmental data on top of the framework Pentaho.

This year we have been working on the efficient computation of OLAP queries related to realistic scenarios proposed by experts in the domain. Precisely, we have devised indexing schemes to access the data in the OLAP cube. We have also worked on the visualization by a GIS (Geographical Information System) of the query results on maps of the geographical area under interest. A paper have been submitted to the COMPAG Journal. This work is detailed in Tassadit Bouadi's thesis [5] .

Efficient computation of skyline queries in an interactive context

Skyline queries retrieve from a database the objects that maximize some criteria, related to user preferences for example, or objects that are the best compromises satisfying these criteria. When data are in huge volumes, such objects may shed light on interesting parts of the dataset. However, computing the skylines (i.e. retrieving the skyline points) may be time consuming because of many dominance tests. This is, especially the case in an interactive setting such as querying a data cube in the context of a datawarehouse.

We have worked at improving the formal setting of the partial materialization of skyline queries when dynamic preferences are refined online by the user. We have explicited which parts of the skyline evolve (which point are added or removed) when a new dimension is introduced in the computation. This led to an efficient incremental method for the online computation of the skyline corresponding to new user preferences [46] . An extended version of this paper is published in Journal "Transactions on Large Scale Data and Knowledge Centered Systems" (TLDKS) [8] and in Tassadit Bouadi's thesis [5] .

Hierarchical skylines

Conventional skyline queries retrieve the skyline points in a context of dimensions with a single hierarchical level. However, in some applications with multidimensional and hierarchical data structure (e.g. data warehouses), skyline points may be associated with dimensions having multiple hierarchical levels. Thus, we have proposed an efficient approach reproducing the effect of the OLAP operators "drill-down" and "roll-up" on the computation of skyline queries. It allows the user to navigate along the dimensions hierarchies (i.e. specialize / generalize) while ensuring an online calculation of the associated skyline. The method is described in Tassadit Bouadi's thesis [5] . A paper describing this contribution is currently under submission to the "Very Large Data Bases (VLDB 2014)" conference.

Modeling influence propagation by Bayesian causal maps

The goal of this project is modeling shellfish fishing to assess the impact of management pollution scenarios on the Rade de Brest. Cognitive maps were built from interviews with fishermen. To represent and reason about these cognitive maps, we propose to use Bayesian Causal Maps making use of fishermen knowledge, particularly to perform influence propagation [82] .

However, this model does not take into account the variety of influences asserted by the fishermen, but only the "mean" causal map. A report describing the project is available [28] . An approach that could combine individual knowledge with belief functions in the way of Philippe Smets's Transferable Belief Model [83] has been proposed. A report describing the projectis available [28] .

This work is done in the framework of the RADE2BREST project, involving Agrocampus Ouest and CNRS (GEOMER/LETG), funded by "Ministère de l'Ecologie" (This project is not mentioned in section 8.2 because DREAM is not an official partner of this project.).

Recommending actions from classification rules

In the framework of the Sacadeau project, a paper dedicated to building actions from classification rules has been published in the KAIS Journal [9] . Our goal is to burden of analysing a large set of classification rules when the user is confronted to an "unsatisfactory situation" and needs help to decide about the appropriate actions to remedy to this situation. The method consists in comparing the situation to a set of classification rules. For this purpose, we propose DAKAR, a new framework for learning action recommendations dealing with complex notion of feasibility and quality of actions.

Sacadeau-Software, which is the decision support tool implemented with F. Ployette (former Inria engineer in the EPI Dream, now retired) in the Sacadeau project, has been published in the RIA Journal [10] . Sacadeau-Software allows to run simulations throughout a watershed and obtain the transfer rate of pollution through the catchment. Classification rules, characterizing the sub-parts of the watershed with pollution and the sub-parts without pollution, are automatically learned from the simulations. A visualization tool enables to relate the learned rules to the examples characterized by these rules. Finally, a user can select a situation of pollution and the action recommendation tool analyses the learned rules and proposes actions that improve this situation of pollution.