Tutorial Presentations:
Epistemic logics for multi-agent systems
Epistemic logic models knowledge and belief in multi-agent systems. How to model change of knowledge has been investigated since the 1980s. Following the influential Interpreted Systems approach as presented in the textbook 'Reasoning about Knowledge' by Fagin et al. (1995), a strand of modal logic with epistemic modal operators for knowledge and dynamic modal operators for knowledge change has developed during the 1990s, based on initial contributions by Plaza (1989) and van Benthem (1987/1989). The action model approach proposed by Baltag, Moss, Solecki (1998) has been the most influential. The textbook 'Dynamic Epistemic Logic' (van Ditmarsch et al, 2007) gives an overview of these developments. The area has continued to develop strongly, for example also incorporating factual change, belief revision, and propositional quantification. In these two tutorial sessions I will give the basics of Dynamic Epistemic Logic, focussing on the semantics and on applications. The topics treated are: logic of knowledge, public announcements, action models, belief revision, factual change, the relation to temporal modal logic (paving the way for the next lecturer), and quantifying over information change.
Game logics
We will introduce the basics of game logic which was proposed by Rohit Parikh in 1985. Then we will have a brief look at the various variants of game and strategy logics that have been studied in the past decade or so, discuss some areas of multi-agent systems where these logics provide a natural way of describing the nuances of the systems. Finally, we will consider some topics/questions for further research in this area.
Approximate reasoning in multi-agent systems
Reasoning about knowledge is one of the most challenging problems in philosophy, artificial intelligence and logic. We focus on rough set philosophy to describe reasoning about knowledge, which is understood as the ability to classify objects. Rough sets are used as a mathematical tool to deal with uncertain and imprecise data. The aim of this talk is to investigate some of the multi-agent knowledge logics with semantics based on the structures inherited from rough set theory.
Planning in multi-agent systems
Planning Domain Definition Language (PDDL) is an attempt to standardize Artificial Intelligence (AI) planning languages. We first discuss how PDDL based languages provide a natural and concise representation for multi agent planning domains. In a multi agent setting, different types of actions are performed by a group of agents. We consider concurrent actions with interacting effects and joint actions. Next we discuss an epistemic logic based planning for multiple agents. It can specify a more complex class of planning domains, specifically it can address nondeterminism, partial observability, and knowledge. Then we briefly discuss an epistemic logic to address trials in planning for multiple agents.
Questions for multi-agent logics
We start with what seem like natural notions of agency and social interaction between agents, and define what seem to be reasonable computational models for multi-agent systems. We discuss the use of modal epistemic logics, temporal logics and game-like logics for reasoning about different aspects of such systems, and point to conceptual difficulties in combining them.
Deontic logics for multi-agent systems
This will be a tutorial on Deontic Logic, the logical study of normative concepts like obligations and permissions. I will start by presenting the system called "Standard Deontic Logic" (SDL), and will sketch its historical development in philosophy, multi-agent systems, and law. Then I will turn to three current issues for deontic logicians: free choice permissions; defeasible obligations and normative conflicts; justification of norms. For each of these I will present both the conceptual questions that they raise, and some logical formalism that have been used to study them. This will include non-normal modal logics, default logic and modal fixed-point logic. The tutorial will be as self-contained as possible. Previous acquaintance with (normal) modal logic will be an asset.
Game theory and dynamics of networks
Modern game theory, whose foundations lie in the intersection of mathematics and economics has become an important tool in computer science. The first lecture is intended to be a tutorial on game theory where we introduce some of the fundamental concepts like dominance, equilibrium and security strategies. We look at the relationship between these concepts and also consider the important notion of improvement paths in games. In the second lecture, we discuss a certain class of network games. We consider a model of network games which incorporates a notion of neighbourhood structures. Each player's choice is influenced by the choice made by other players in this neighbourhood structure. Existence of a stable strategy profile is a fundamental issue in such a setting. We also discuss the question of whether stable profiles can be computed as convergence of some update dynamics.
Planning with epistemic goals
In this talk, we survey recent approaches to planning with epistemic goals in single and multi-agent settings. In particular, we focus on the work based on dynamic-epistemic logics. We point out some conceptual and technical difficulties, and suggest possible future directions which may shed new light on AI planning in general.