The benefits of forgetting knowledge

PhD position / Sujet de thèse

Agents elaborate their knowledge along time. It changes depending on their experience and communication with other agents. These changes are often considered as a never-ending improvements. However, there are circumstances in which it may be worth to forget part of this knowledge. This may be for reason of efficiency or boundedness, of changes in the world in which agents live or of of change in the knowledge itself. The goal of this proposal is to study forgetting mechanisms in the context of cultural knowledge evolution.

Cultural knowledge evolution deals with the evolution of knowledge representation in a group of agents. For that purpose, cooperating agents adapt their knowledge to the situations they are confronted to and the feedback they receive from others. This framework has been considered in the context of evolving natural languages [Steels, 2012]; We have applied it to ontology alignment repair, i.e. the improvement of incorrect alignments [Euzenat, 2014; 2017]. We have shown that cultural repair is able to converge towards successful communication through improving the objective knowledge quality.

So far, most of the work has concerned the way agents acquire knowledge either directly from the environment or indirectly by communicating with other agents. We want to consider the possibility for agents to forget knowledge, or more exactly to revise beliefs. This may occur for several reasons: because this knowledge has been proved invalid, because it is not valid, or useful, anymore, or because agents have limited memory capability and must arbitrate between what they know. These three categories roughly cover revision [Alchourrón, 1985], update [Katsuno, 1991] and abstraction [Euzenat, 1991]. An interesting issue resides in the dynamics of the processes over time to choose among these categories the one to be executed in a given context.

The Sars-CoV-2 pandemic could be taken as an inspiration source: fast scientific knowledge acquisition and administrative regulation changes have required agents to revise their 'knowledge'. In principle, scientific knowledge is considered as a constantly growing corpus. However, the fast dissemination of non-safely established or insufficiently narrowed results, required the revision of what people thought. Subsequent or independent regulation changes required them to update their behaviour accordingly. Another source of inspiration could be taken in the evolution of social network to assist people in revising their links.

We propose to study the costs and benefits of forgetting knowledge. This requires measuring the impact of these various forms of forgetting on the behaviour and knowledge elaborated by agent societies. This impact may be assessed through measures of the success rate or understanding of agents within a society or the correctness, efficiency and size of the resulting knowledge. For that purpose, it is necessary to design adequate forgetting mechanisms and to establish their properties either experimentally —through multi-agent simulation— or theoretically —through formal modelling and proof. We already have available frameworks for simulating the cultural evolution of knowledge and modelling it. Such frameworks may be extended. A particular attention will be given to the modelling and in the understanding the dynamics of the knowledge and of the mechanisms over time, as already explored in [Lacomme, 2009; 2011].

This work is part of an ambitious program towards what we call cultural knowledge evolution partly funded by the MIAI Knowledge communication and evolution chair.


[Alchourrón, 1985] Carlos Alchourrón, Peter Gärdenfors, David Makinson, On the logic of theory change: partial meet contraction and revision functions, Journal of symbolic logic 50(2):510-530, 1985
[Euzenat, 1991] Jérôme Euzenat, Libero Maesano, An architecture for selective forgetting, Proc. 8th AISB conference, Leeds (UK), pp117-128, 1991
[Euzenat, 2014] Jérôme Euzenat, First experiments in cultural alignment repair (extended version), in: Proc. 3rd ESWC workshop on Debugging ontologies and ontology mappings (WoDOOM), Hersounisos (GR), LNCS 8798:115-130, 2014
[Euzenat, 2017] Jérôme Euzenat, Communication-driven ontology alignment repair and expansion, in: Proc. 26th International joint conference on artificial intelligence (IJCAI), Melbourne (AU), pp185-191, 2017
[Katsuno, 1991] Hirofumi Katsuno, Alberto Mendelzon, On the difference between updating a knowledge base and revising it, Proc. 2nd international conference on principles of knowledge representation and reasoning, Cambridge (MA US), 1991
[Lacomme, 2009] Laurent Lacomme, Yves Demazeau, Valérie Camps, Classification des mécanismes organisationnels dans les réseaux d’agents, 17ème Journées Francophones sur les Systèmes Multi-Agents (JFSMA), Guessoum, Hassas eds, pp. 79-88, Cepadues, Lyon, Octobre 2009.
[Lacomme 2011] Laurent Lacomme. Une modèle générique pour les organisations dynamiques en univers multi-agent. Thèse de doctorat de l’Université de Grenoble. 2011.
[Steels, 2012] Luc Steels (ed.), Experiments in cultural language evolution, John Benjamins, Amsterdam (NL), 2012


Qualification: Master or equivalent in computer science.

Researched skills:

Doctoral school: MSTII, Université Grenoble Alpes.

Advisor: Jérôme Euzenat (Jerome:Euzenat#inria:fr) and Yves Demazeau (Yves:Demazeau#imag:fr).

Group: The work will be carried out in the mOeX team common to INRIA & LIG with the support of Yves Demazeau (CNRS) also at LIG. mOeX is dedicated to study knowledge evolution through adaptation. It gathers researchers which have taken an active part these past 15 years in the development of the semantic web and more specifically ontology matching and data interlinking. Yves Demazeau is one of the founders of the multi-agent research domain in Europe and has more than 30 years’ experience in this domain.

Place of work: The position is located at INRIA Grenoble Rhône-Alpes, Montbonnot a main computer science research lab, in a stimulating research environment.

Hiring date: as soon as possible.

Duration: 36 months

Salary: 1760€/month (gross, before social contributions and taxes).

Deadline: as soon as possible.

Contact: For further information, contact us.

Procedure: Contact us.

File: Provide Vitæ, motivation letter and references. It is very good if you can provide a Master report and we will ask for your marks in Master, so if you have them, you can join them.