Section: New Results

New results on Axis 4: Data-driven Models

Crowdsourcing

Participants : Loïc Hélouët, Rituraj Singh

Complex workflows for crowdsourcing.

Crowdsourcing consists in hiring workers on internet to perform large amounts of simple, independent and replicated work units. We have proposed [32] complex workflows, a model for concurrent orcestration of tasks to solve problems that are more intricate than simpe tagging problems. Complex workflows allow higher-order answers where workers can suggest a process to obtain data rather than a plain answer. It is a data-centric model based on orchestration of concurrent tasks and higher order schemes. We have considered formal properties of specifications described with this model termination (whether some/all runs of a complex workflow terminate) and correctness (whether some/all runs of a workflow terminate with data satisfying FO requirements). We have shown that existential termination/correctness are undecidable in general excepted for specifications with bounded recursion. However, universal termination/correctness are decidable when constraints on inputs are specified in a decidable fragment of FO, and are at least in 2EXPTIME.

CrowdInc : a solution to reduce the cost of Consensus in Crowdsourcing.

Another contribution around crowdsourcing [34] considers agregation of answers, reliability of computed results, and optimization of costs. Crowdsoucing call for human expertise to solve problems which are still hard for computers, but easy for human workers. Crowdsourcing platform distribute replicated tasks to workers, pay them for their contribution, and aggregate answers to produce a reliable conclusion. A fundamental problem is to infer a correct answer from the set of results returned by workers. An additional ingredient of crowdsourcing is the cost needed to obtain a reliable answer: unlimited budget allows for the use of large pools of workers for each task, or experts to improve reliability of aggregated answers, but a limited budget forces to use resources at best to synthesize an reasonably reliable answer. We have focused on crowdsourcing of simple tasks with boolean answers. In this setting, we have first defined a probabilistic inference technique to agregate answers. This allows to consider difficulty of tasks and expertise of workers when building a conclusion. We have then proposed a greedy algorithm that reduces the cost (i.e. the number of workers hired by a platform) needed to reach a consensual answer. This algorithm considers difficulty of task, budget provided by client and total tasks to dynamically adapt threshold at each stage and makes locally optimal choice while preserving accuracy. Last, we have shown efficiency of our algorithm on several benchmarks, and compared its performance to existing solutions.

Guarded Attribute Grammars and Petri net synthesis

Participants : Adrian Puerto Aubel, Éric Badouel

Service-oriented programming

We addressed [30] the problem of component reuse for the design of user-centric distributed collaborative systems modelled by Guarded Attribute Grammars. Following the contract-based specification of components we develop an approach to an interface theory for the components of a collaborative system in three stages: we define a composition of interfaces that specifies how the component behaves with respect to its environement, we introduce an implementation order on interfaces and finally a residual operation on interfaces characterizing the systems that, when composed with a given component, can complement it in order to realize a global specification.

The visit of Joskel Ngoufo, a doctoral student at Yaoundé University, was the occasion to initiate a new implementation of the Guarded Attribute Grammars engine, in Racket language, a dialect of Lisp that allows metalanguage facilities and graphical interfaces to be processed more easily than in Haskell, the language chosen for the previous implementation.

Coordination of public debate.

Our research on data-centric collaborative systems has focused this year on the modelling of debates [28], with the aim of producing a tool that makes it possible to automatically conduct them, while managing relevant documents and analysing the respective positions of the different interventions from the point of view of argumentation theory. To this end, we are collaborating with Carlo Ferigato, a researcher at the JRC (C.E. Ispra, Italy), an institute for which we jointly produced a report covering an overview of the different theories developed around the subject, as well as the main tools proposing solutions to this problem. The aim of this collaboration is at understanding the basic principles and the computer programs apt to coordinate a public debate with an overall aim at giving the bases for designing such programs. Computer programs for the coordination of public debate exist since the beginning of the eighties but recently they have acquired new relevance for the use made of them by public administrations, associations and political parties. The meet of both citizen's needs and public administrations for transparency can today be technically realized with such programs through the present communication means in a more efficient way with respect to the first experiments dating now about forty years. This report aims at covering historical, technical and some theoretical aspects of the use of computers for the coordination of public debate.

Orthomodular partial orders.

The collaboration with Carlo Ferigato, is in line with the latter's thesis subject [11]. The set of regions of a condition/event transition system represents all the possible local states of a net system the behaviour of which is specified by the transition system. This set can be endowed with a structure, so as to form an orthomodular partial order. Given such a structure, one can then define another condition/event transition system. We study cases in which this second transition system has the same collection of regions as the first one. When it is so, the structure of regions is called stable. We proposed, to this aim, a composition operation, and a refinement operation for stable orthomodular partial orders, the results of which are stable.