PANAMAGENTS: The Agents Seminar in Paris

Seminars located in LIPADE
PANAMAGENTS is a seminar on intelligent agents and multi-agent systems, the offspring of a collaboration among three Parisian labs: LIPADE/DAI group (Univ. Paris-Descartes), LIP6/SMA Group (Univ. Pierre et Marie Curie), and LAMSADE (Univ. Paris Dauphine). Each seminar is taking place in one of these labs.


Back to DAI's webpage

Start

2015

  • 07/12/15--Jaime Simão Sichman (University of São Paulo, Brazil): Improving the performance of BDI Jason agents through perception filters

    Abstract: When agents are supposed to be integrated to virtual environments virtual or simulators, one of the BDI paradigm’s major concerns is the lack of control over the agents’ perceptions. Without having any form of goal directed perceptions, the agent may be flooded by irrelevant information thus causing an unjustified increase in processing time. In order to provide greater control on the agent’s perceptions and to reduce its time response, this work presents a filtering perception me- chanism for the Jason interpreter, aimed at eliminating those perceptions that can be ignored. To this end, some types of pre-defined filters have been proposed, implemented, and applied to three different scenarios. Through appropriate statistical validation methods, it was shown that applying perception filters can reduce up to 80 % of an agent’s processing time, without significantly affecting its performance measured in terms of its utility function.

2014

  • 30/09/14--Marie desJardins (Univ. Maryland, Baltimore, USA): Subgoal Discovery and Language Learning in Reinforcement Learning Agents

    Abstract: As intelligent agents and robots become more commonly used, methods to make interaction with the agents more accessible will become increasingly important. In this talk, I will present a system for intelligent agents to learn task descriptions from linguistically annotated demonstrations, using a reinforcement learning framework based on object-oriented Markov decision processes (OO-MDPs). Our framework learns how to ground natural language commands into reward functions, using as input demonstrations of different tasks being carried out in the environment. Because language is grounded to reward functions, rather than being directly tied to the actions that the agent can perform, commands can be high-level and can be carried out autonomously in novel environments. Our approach has been empirically validated in a simulated environment with both expert-created natural language commands and commands gathered from a user study. I will also describe a related, ongoing project to develop novel option discovery methods for OO-MDP domains. These methods permit agents to identify new subgoals in complex environments that can be transferred to new tasks. We have developed a framework called Portable Multi-policy Option Discovery for Automated Learning (P-MODAL), an approach that extends the PolicyBlocks option discovery approach to OO-MDPs. This work is collaborative research with Dr. Michael Littman and Dr. James MacGlashan of Brown University, Dr. Smaranda Muresan of Columbia University. A number of UMBC students have contributed to the project: Shawn Squire, Nicholay Topin, Nick Haltemeyer, Tenji Tembo, Michael Bishoff, Rose Carignan, and Nathaniel Lam.


  • 24/11/14-- John Mylopoulos (University of Trento, Italy): Designing Adaptive Software Systems: A Requirements Engineering Perspective

    Abstract: Adaptive systems usually operationalize adaptation through a feedback loop, an architectural prosthesis that introduces monitoring, diagnosis and compensation functions to the system proper. We have been studying the requirements that lead to such feedback loop functionality. In particular, we have introduced new classes of requirements, called respectively awareness and evolution requirements, which are best operationalized through feedback loops instead of collections of functions. These requirements are characterized by the fact that they refer to other requirements, quality constraints or domain assumptions, and their success or failure. We then discuss elicitation, modeling, formalization for awareness and evolution requirements and how to go from such requirements to feedback loops through a systematic process. In addition, we sketch a framework for monitoring, diagnosis and compensation grounded on requirements models. This is joint work with Vitor Souza (UFES Brazil), Kostas Angelopoulos (UniTN Italy) and Alexei Lapouchnian (UToronto Canada).

2013

  • 01/02/13--Mark KLEIN (MIT, USA): Enabling Large-Scale Deliberation About Complex Problems

    Abstract: Humanity now finds itself faced with highly complex challenges – ranging from climate change and the spread of disease to product development and scientific collaborations - that require effective decision making with large diverse communities that are distributed in time and space. While social computing tools (e.g. web forums, wikis, email, instant messaging, media sharing sites, social networking, and so on) have created unprecedented opportunities for connecting and sharing on a massive scale, they still fare poorly when applied to decision making with complex controversial problems. Internet-mediated discussions are instead all-to-often characterized by very poor signal-to-noise ratios, widely varying quality, spotty coverage, scattered content, and other serious weaknesses. This talk will describe two interconnected projects aimed at addressing these important challenges: 1) the Deliberatorium: a large-scale argumentation approach designed to help groups quickly enumerate the space of possible solutions 2) nonlinear negotiation: protocols designed to help stakeholders find win-win solutions from within such large solution spaces. I will describe the underlying concepts for these projects, the results of our evaluations to date, and some promising directions for future work.


  • 9/4/13--Carles SIERRA (Institute of Research on Artificial Intelligence of the Spanish Council for Scientific Research (CSIC)) : Information-based trust and reputation

    Abstract: In this talk I will describe how information-based techniques can be used to underpin an expressive and practical trust and reputation model. The model distinguishes between objective and subjective opinions that are then aggregated to build up a reputation measure. The architectural requirements for agents using this model will be discussed and two use cases: eProcurement and scientific publications management, will illustrate the use of the model.

2012

  • 24/10/12--Antonis KAKAS (Univ. of Cyprus): Argumentation Logic

    Abstract: This talk will present a new Logic of Arguments that is close to human reasoning and to the original link, dating back to ancient Greek philosophers, between logic and human argumentation. This new logic, called Argumentation Logic, is developed by bringing together elements from argumentation theory in Artificial Intelligence and the syllogistic view of classical logic as provided by its Natural Deduction proof system. It constitutes an argumentation based reformulation of Classical (Propositional) Logic where logical formulae are regarded as arguments and logical conclusions are given under a suitable criterion of acceptability of arguments. Argumentation Logic can be shown to be equivalent to classical Propositional Logic for classically consistent propositional theories. Importantly, the logic does not trivialize under inconsistent information but rather isolates out the inconsistency so that conclusions whose supporting arguments are not affected by the inconsistent information can still be drawn. The talk will also briefly discuss the potential significance of such an argumentation perspective of logic for the new paradigm of computing that is emerging today out of many applications over the World Wide Web where there is a need for more human-like decision making and interaction amongst artificial and human entities.


  • 26/10/12--Francesca Toni (Imperial College, UK): Mechanism Design for Argumentation-based Persuasion

    Abstract: Recently we have seen a few development in argumentation-based dialogue systems, but there is less research in understanding agents’ strategic behaviour in dialogues. We study agent strategies by linking a specific form of argumentation-based dialogues and mechanism design. Specifically, ocusing on persuasion dialogues, we show how dialogues can be mapped to concepts in mechanism design. We prove that a “truthful” and “thorough” dialogue strategy is a dominant strategy under specific conditions. We also prove that a mechanism using this dialogue strategy implements a “persuasion social choice function” we define. These results show the validity of the proposed strategies for gents in persuasion and the feasibility of studying persuasion with mechanism design techniques.


  • 31/10/12--Francesca Toni (Imperial College, UK): Argumentation and the Web

    Abstract: I will provide an overview of computational argumentation, focusing on abstract argumentation and assumption-based argumentation, as well as uses of these forms of argumentation in Web contexts, and in particular Semantic Web as well as Social Networks contexts. I will outline achievements to date as well as open issues and challenges.

2011

  • 19/05/11--Alessio Lomuscio (Imperial College London, UK): Verification of multi-agent systems

    Abstract: Multi-agent systems are distributed autonomous systems in which the components, or agents, act autonomously or collectively in order to reach private or common goals. Logic-based specifications for MAS typically do not only involve their temporal evolution, but also other intensional states, including their knowledge, beliefs, intentions and their strategic abilities. This talk will survey recent work carried out on model checking MAS. Specifically, serial and parallel algorithms for symbolic model checking for temporal-epistemic logic as well as bounded-model checking procedures will be discussed. MCMAS, an open-source model checker, developed at Imperial College London, will be briefly demonstrated. Applications of the methodology to the automatic verification of security protocols, web services, and fault-tolerance will be surveyed.


  • 30/06/11--Brahim Chaib-draa (Université de Laval (Québec) Canada) : Apprentissage et planification multi-agents

    Abstract: Prendre les bonnes décisions dans des environnements multi-agents est une tâche difficile dans la mesure où la présence de plusieurs décideurs implique des conflits d'intérêts, un manque de coordination, et une multiplicité de décisions possibles. Si de plus, les décideurs interagissent successivement à travers le temps, ils doivent non seulement décider ce qu'il convient de faire actuellement, mais aussi comment leurs décisions actuelles peuvent affecter le comportement des autres dans le futur. La théorie des jeux est un outil mathématique qui vise à modéliser ce type d'interactions via des jeux stratégiques à plusieurs joueurs. Des lors, les problèmes de décision multi-agents sont souvent étudiés en utilisant la théorie des jeux. Dans ce contexte, et si on se restreint aux jeux dynamiques, les problèmes de décision multi-agents complexes peuvent être approchés de façon algorithmique en utilisant de tels jeux. Partant de ces considérations, je ferai le tour de trois contributions. La première est un cadre algorithmique pour la planification distribuée dans les jeux dynamiques non-coopératifs. Sans une telle planification distribuée, la multiplicité des plans possibles pourrait générer de graves complications pour la planification. Dans ce contexte, nous avons proposé une nouvelle approche basée sur l'apprentissage dans les jeux répétés. Une telle approche permet de surmonter lesdites complications par le biais de la communication entre les joueurs. La deuxième contribution est un algorithme d'apprentissage pour les jeux répétés en «self-play». Cet algorithme permet aux joueurs de converger, dans les jeux répétés initialement inconnus, vers un comportement conjoint optimal dans un certain sens bien défini, et ce, sans aucune communication entre les joueurs. La troisième contribution est une famille d'algorithmes de résolution approximative des jeux dynamiques et d'extraction des stratégies des joueurs. Cette famille comprend une méthode pour calculer un sous-ensemble non vide des équilibres approximatifs parfaits en sous-jeu dans les jeux répétés. Cette méthode est ensuite étendue pour approximer tous les équilibres parfaits en sous-jeu dans les jeux répétés, permettant ainsi de résoudre des jeux dynamiques plus complexes. Finalement, je ferai état de quelques problèmes ouverts que je pourrais faire en collaboration avec les collègues de Paris Descartes.

Before

  • 08/10/09--John Mylopoulos (Univ. of Trento, IT): Agent-Oriented Software Engineering: A Requirements Engineering Perspective

    Abstract: The last fifteen years have seen the rise of a new phase in software development which is concerned with the acquisition, modelling and analysis of stakeholder purposes ("goals") in order to derive functional and non-functional requirements. This phase is founded on the concepts of goal, actor as well as inter-actor dependencies. We review the history of ideas and research results for this new phase and sketch on-going research on the topic. Specifically, we discuss the Tropos methodology that uses these concepts to support all phases of software development. The presentation includes some of our on-going work on Tropos. The research reported is the result of collaborations with colleagues at the Universities of Toronto and Trento.


  • 04/04/08--Ronen Brafman, Ben Gurion University: Efficient Planning for Loosely Coupled Multi-Agent Systems

    Abstract: Loosely coupled multi-agent systems are perceived as easier to plan for because they require less coordination between agent sub-plans. In this talk I will attempt to formalize this intuition. First, this requires us to establish some measure of the coupling level of a system. I will suggest two parameters, one that is system dependent and one that is problem dependent. I will then show how planning complexity is influenced by these parameters. The key property of this result is that there is no direct dependence on the number of agents in the system. That is, if the number of agents increases by the level of coupling remains fixed, planning complexity scales up polynomially.


  • 16/10/07--Katia SYCARA, Carnegie Mellon (USA): The Many Faces of Automating and Supporting Multi Agent Interactions

    Abstract: Software agents represent a radical departure from earlier monolithic approaches to decision support by introducing intelligence in small packages in many different places. The robustness of distributed intelligence could offer many advantages in supporting humans in different types of interactions. In this talk we will discuss our research in modeling autonomous interactions of coordinating agents in different coordination regimes (e.g. negotiation, coalition formation) and in different settings, especially in open, dynamic and large scale environments. Additionally, we will present our work on providing agent-based support to interacting humans. Agents could be very useful in supporting human teamwork in the following roles: (a) an agent could support the task of an individual team member, (b) an agent could play the role of a team mate, and (c) an agent can support the team as a whole. Such agent support has the benefits that the human team could offload tasks to agents to reduce cognitive load, or the agents could aid in monitoring or synchronizing team activity.