Laboratoire d'informatique de l'École polytechnique

Sémin'Ouvert : « Semantic Structures for Spatially-Distributed Multi-Agent Systems » par Frank Valencia

Speaker: Frank Valencia (CNRS, LIX École polytechnique)
Location: Bât. Turing, amphi. Germain
Date: Jeu. 19 oct. 2017, 14h30-15h30

My research in COMETE involves the study of formal languages and semantic structures for concurrent systems and their relationship to logic. In this talk I shall introduce an algebraic structure, which we call SCS, for specifying the behaviour of spatially-distributed multi-agent systems. In the systems under consideration agents have spaces (local stores) where they may have epistemic information (e.g., facts, beliefs, opinions, "alternative" facts), processes and other spaces. Agents communicate by using processes that can move epistemic information and other processes across different spaces. SCS's can be seen as Scott's information systems with additional structure for specifying agents and their spaces. From a computational point of view scs can be used to specify partial information holding in a given agent's space (local information). From an epistemic point of view SCS can be used to specify information that a given agent considers true (beliefs). They can also specify the mobility of information/processes from one space to another. They also provide for process/information extrusion, a central concept in formalisms for mobile communication from concurrency theory. From an epistemic point of view extrusion corresponds to a notion we shall call utterance; a piece of information that an agent communicates to others but that may be inconsistent with the agent’s beliefs. Finally, scs can also also distributed information, understood as information that agents may conclude if they were to combine their local information. SCS have been used as semantics structures for modal logic and spatial timed reactive programming languages. In this presentation I shall talk about SCS with emphasis on recent formal developments such as the notion of distributed information and a prototype of an online multi-agent platform (DSpaceNet) where users (agents) interact much like in social networks. The distinct feature of DSpaceNet is that not only can the users post messages in their spaces but also mobile programs that can migrate across different spaces. Thus users can for example write and post their own watchdog processes that react to certain messages. This presentation is based on joint work with Michell Guzman, Salim Perchy, Sophia Knight, Prakash Panangaden, Camilo Rueda, Stefan Haar and Hector Delgado. The talk does not presuppose knowledge about concurrency theory.