2019-01506 - Doctorant F/H CORDIS2019-EPIGraphIK-Explanation for Defeasible Reasoning

Type de contrat : CDD de la fonction publique

Niveau de diplôme exigé : Bac + 5 ou équivalent

Fonction : Doctorant

Contexte et atouts du poste

Les travaux de l'équipe se situent dans le domaine de la représentation des connaissances et des raisonnements. L'équipe GraphIK a une approche  computationnelle de ce domaine, basée sur la logique et également sur des  graphes étiquetés, ce qui constitue leur angle d'attaque particulier. Ils s'intéressent à quelques uns des problèmes importants de ce domaine comme  l'interrogation de grandes bases de connaissances, le raisonnement dans des bases de connaissances hybrides c'est-à-dire dont les composants ont chacun leurs propres mécanismes de raisonnement, le traitement de connaissances  "imparfaites".

voir descriptif en anglais

Mission confiée

 

Context & motivating scenario

In logic, defeasible reasoning frameworks [1] allow to reach a conclusion
to be later withdrawn in the presence of new evidence that proves
otherwise. Statement graphs [2]encode defeasible reasoning
using labeled graphs. As proven in [2], labeling functions allow soundness and completeness wiith respect to the semantics of defeasible logic. Explanation is a term widely used in the context of knowledge base
reasoning typically aimed at clarification (i.e. assuring to the user that
the reasoning behind its conclusions is sound) and/or persuasion (i.e.
convincing the user that system's conclusion is appropriate as opposed
to another). In this thesis we will focus on clarification explanations that are concerned with the structure of logical formulae
that supports the status of a query with respect to a defeasible knowledge base. 

Shortcomings of the state of the art

Existing work on explanations in knowledge based systems focused on the representation of the structure used for explaining. However, none of the existing works investigated such structures in the context of defeasible reasoning. 
In our case, we will turn our attention to statement graphs as our
explanation structure. Such approach has the fundamental drawback
of graph size. As the graph's knowledge base grows larger, understanding
it becomes harder, thus affecting the clarification capabilities. 

Proposed research program: prioritized research questions, methods, evaluation techniques

 
In order to address this research question one should answer several research questions shown below, in prioritised order: What is the added value of using Statement graphs over the existing defeasible reasoning formalisms?, How can one rearrange and/or transform a statement graph in expansion steps in order to provide clear and minimal explanations?, How can one prioritize and group the information in the expansion steps? and How to take into account the user and its specificity for the explanation process?.
 

 

Principales activités

Therefore, the thesis will follow the following research plan (to be eventually updated in function of the outcomes of the intermediate steps):
 
1. State of the art research into defeasible logics formalisms (expressivity, reasoning mechanisms, practical applicability, existing implementations) (6 months)
2. Research into the different mechanisms for expansion to be used for Statement Graphs (DFS, BFS, etc.) (6 months)
3. Research into ranking semantics for Statement Graphs (computation, soundness, desirable properties etc.) (6 months)
4. Research into how ranking semantics can be combined with expansion mechanisms (6 months)
5. Expanding the ranking semantics developed into the previous step to explore how the user interaction could provide basis for such mechanism. Experimentation to confirm / infirm the intuitions (6 months)
6. Revise the ranking at step 5 according to the experiment outcome and provide a final mechanism.
Final experimentation (6 months).

 

Compétences

Technical skills and level required : excellent knowledge of logic and programming

Languages : excellent knowledge of English

Relational skills : independent and keen to work in a team

Avantages

  • Restauration subventionnée
  • Transports publics remboursés partiellement
  • Congés: 7 semaines de congés annuels + 10 jours de RTT (base temps plein) + possibilité d'autorisations d'absence exceptionnelle (ex : enfants malades, déménagement)
  • Possibilité de télétravail (après 6 mois d'ancienneté) et aménagement du temps de travail
  • Équipements professionnels à disposition (visioconférence, prêts de matériels informatiques, etc.)
  • Prestations sociales, culturelles et sportives (Association de gestion des œuvres sociales d'Inria)
  • Accès à la formation professionnelle
  • Sécurité sociale