MAD

The Models theme focuses on time and space, preferences, and reasoning algorithms. We are interested in knowledge representation with logics, reasoning, with a particular attention paid to the representation of time and space on the one hand, and to the expressivity and algorithmics of logical fragments, on the other hand. We mainly develop formal models, for applications, and algorithms or proofs of complexity.

The Models theme deals for example with the interaction with the human being, by studying languages from the point of view of their “natural” aspect: ability to represent temporal expressions with different granularities (seconds, days, months), to convert from one to the other, or to represent fuzzy or imprecise expressions (“after dinner”). Interaction with human beings is also approached from a formal and systematic angle, in addition to work carried out on specific applications: for example, in the context of a multidisciplinary ANR project on risks in socio-economic development, or for the representation of events in a context of knowledge extraction from documents, in the context of several industrial PhD. theses with Airbus Defence & Space.

The Agents theme focuses on the formal specification and validation of autonomous agents, and on trust and manipulations in multi-agent systems. Hence, we are interested in multi-agent systems, seen both as tools that are designed and implemented, and as objects of analysis, especially in a context of strategic interactions. The tools are formal (game theory, theorem proof) and experimental (simulation).

For instance, we use theorem proof as an original approach for validation. The aim is to propose formal languages in which to describe potentially interacting agent behaviours, and tools to prove that systems implementing these specifications achieve given goals (Goal Decomposition Trees). From a systems analysis perspective, the Agents theme studies the strategic interactions between agents in decentralized systems: reputation and trust, coalition building. The approach developed here mixes formal analysis (game theory, learning) with simulation. In particular, the team has obtained results on the analysis of reputation protocols and coalition formation mechanisms, and their manipulation by malicious agents. Finally, the Agents theme is also interested in ethical issues, aiming at defining how a system composed of one or more artificial agents can be able to manage ethical conflicts and explicitly justify their choices (ETHICAA ANR project).

The Decision theme focuses on Markov decision processes (MDPs) and their extensions for autonomous decision making in uncertain environments. We focus on algorithmics, and give importance to implementation and concrete experiments. We are interested in the resolution of MDPs, which model decision making problems in uncertain (stochastic) environments, and their extensions to partially observed environments (POMDP) or to several cooperative agents (DEC-MDP and DEC-POMDP). The team has developed new techniques for solving decentralized versions, and implemented the models and algorithms in a number of concrete realizations.

For example, the team has used the DEC-POMDP models and algorithms for problems of mapping unknown areas, with multiple agents and without (explicit) communication at runtime. This work led to a real use case on robots, within the framework of an ANR/DGA (Robots Malins) challenge, for which the team was again vice-champion in 2012 (after 2011). Markovian models are also used in the framework of several CIFRE theses in collaboration with AirbusDefence & Space, for problems related to the detection of group behaviour, threat management and information gathering via sensor networks. Finally, the European project COACHES, led by the team, proposed to implement these techniques for human-service robots, whose deployment has been tested in a shopping center. Beyond the decision making problems, it raised issues in order to take into account the human in the decision-making process.

BONNET Grégory – University lecturer at the University of Caen Normandie

BOURDACHE Nadjet – University lecturer at the University of Caen Normandie

BOUZID Maroua – Professor of universities at the University of Caen Normandie

JEANPIERRE Laurent – University lecturer at the University of Caen Normandie

MERMET Bruno – University lecturer at the University Le Havre Normandie

MOUADDIB Abdel-Illah – Professor of universities at the University of Caen Normandie

NIVEAU Alexandre – University lecturer at the University of Caen Normandie

SIMON Gaële – University lecturer at the University Le Havre Normandie

ZANUTTINI Bruno – Professor of universities at the University of Caen Normandie

EpiRL (ANR, 2023-2026)

PING/ACK (ANR, 2019-2022)

ONE4YOU (Région Normandie, 2019-2021)

SICoPaD (Normandy region , 2018-2021)

PepTraq (Normandy region, 2017-2019)

ETHICAA (ANR CONTINT, 2014-2018)

COACHES (ERC Chist-ERA, 2014-2017)

GARDES (ANR/DGA, 2014-2017)

ROBOTS_MALINS (ANR, 2009-2012)