A simple asset pricing model with two types of adaptively learning traders, fundamentalists and technical analysts, is studied. Fractions of these trader types, which are both boundedly rational, change over time according to evolutionary learning, with technical analysts conditioning their forecasting rule upon deviations from a benchmark fundamental. Volatility clustering arises endogenously in this model. Two mechanisms are proposed as an explanation. The first is coexistence of a stable steady state and a stable limit cycle, which arise as a consequence of a so-called Chenciner bifurcation of the system. The second is intermittency and associated bifurcation routes to strange attractors. Both phenomena are persistent and occur generically in nonlinear multi-agent evolutionary systems. (author's abstract) / Series: Working Papers SFB "Adaptive Information Systems and Modelling in Economics and Management Science"

In this thesis, we explore the topic of designing an architecture and processing algorithms for a multi-agent system, where agents need to address potential unexpected events in the environment, operating under soft real-time constraints. We first develop a classification of unexpected events into Opportunities, Barriers and Potential Causes of Failure, and outline the interaction required to support the allocation of tasks for these events. We then propose a hybrid architecture to provide for agent autonomy in the system, employing a central coordinating agent. Certain agents in the community operate autonomously, while others remain under the control of the coordinating agent. The coordinator is able to determine which agents should form teams to address unexpected events in a timely manner, and to oversee those agents as they perform their tasks. The proposed architecture avoids the overhead of negotiation amongst agent teams for the assignment of tasks, a benefit when operating under limited time and resource constraints. It also avoids the bottleneck of having one coordinating agent making all decisions before work can proceed in the community, by allowing some agents to work independently. We illustrate the potential usefulness of the framework by describing an implementation of a simulator loosely based on that used for the RoboCup Rescue Simulation League contest. The implementation provides a set of simulated computers, each running a simple soft real-time operating system. On top of this basic simulation we implement the model described above and test it against two different search-and-rescue scenarios. From our experiments, we observe that our architecture is able to operate in dynamic and real-time environments, and can handle, in an appropriate and timely manner, any unexpected events that occur. We also comment on the value of our proposed approach for designing adjustable autonomy multi-agent systems and for specific environments such as robotics, where reducing the overall level of communication within the system is crucial.

Multi-agent systems represent a group of agents that cooperate to solve common tasks in a dynamic environment. Multi-agent control systems have been widely studied in the past few years. The control of multi-agent systems relates to synthesizing control schemes for systems which are inherently distributed and composed of multiple interacting entities. Because of the wide applications of multi-agent theories in large and complex control systems, it is necessary to develop a framework to simplify the process of developing control schemes for multi-agent systems. <br /><br /> In this study, a framework is proposed for the distributed control and coordination of multi-agent systems. In the proposed framework, the control of multi-agent systems is regarded as achieving decentralized control and coordination of agents. Each agent is modeled as a Coordinated Hybrid Agent (CHA) which is composed of an intelligent coordination layer and a hybrid control layer. The intelligent coordination layer takes the coordination input, plant input and workspace input. After processing the coordination primitives, the intelligent coordination layer outputs the desired action to the hybrid layer. In the proposed framework, we describe the coordination mechanism in a domain-independent way, as simple abstract primitives in a coordination rule base for certain dependency relationships between the activities of different agents. The intelligent coordination layer deals with the planning, coordination, decision-making and computation of the agent. The hybrid control layer of the proposed framework takes the output of the intelligent coordination layer and generates discrete and continuous control signals to control the overall process. In order to verify the feasibility of the proposed framework, experiments for both heterogeneous and homogeneous Multi-Agent Systems (MASs) are implemented. In addition, the stability of systems modeled using the proposed framework is also analyzed. The conditions for asymptotic stability and exponential stability of a CHA system are given. <br /><br /> In order to optimize a Multi-Agent System (MAS), a hybrid approach is proposed to address the optimization problem for a MAS modeled using the CHA framework. Both the event-driven dynamics and time-driven dynamics are included for the formulation of the optimization problem. A generic formula is given for the optimization of the framework. A direct identification algorithm is also discussed to solve the optimization problem.

This thesis relates to the usage of trust modelling in multi-agent systems - environments in which there are interacting software agents representing various users (for example, buyers and sellers exchanging products and services in an electronic marketplace). In such applications, trust modelling may be crucial to allow one group of agents (in the e-commerce scenario, buyers) to make effective decisions about which other agents (i.e., sellers) are the most appropriate partners. A number of existing multi-agent trust models have been proposed in the literature to help buyers accurately select the most trustworthy sellers. Our contribution is to propose several modifications that can be applied to existing probabilistic multi-agent trust models. First, we examine how the accuracy of the model can be improved by limiting the network to a portion of the population consisting of the most trustworthy agents, such that the less trustworthy contributions of the remaining agents can be ignored. In particular, we explore how this can be accomplished by either setting a maximum size for a buyer's advisor network or setting a minimum trustworthiness threshold for agents to be accepted into that advisor network, and develop methods for appropriately selecting the values to limit the network size. We demonstrate that for two models, both the Personalized Trust Model (PTM) developed by Zhang as well as TRAVOS, these approaches will yield significant improvements to the accuracy of the trust model, as opposed to using an unrestricted advisor network. Our final proposed modification is to use an advisor referral system in combination with one of the network-limiting approaches. This would ensure that if a particular agent within the advisor network had not met a specified level of experience with the seller under consideration, it could be replaced by another agent that had greater experience with that seller, which should in turn allow for a more accurate modelling of the seller's trustworthiness. We present a particular approach for replacing advisors, and show that this will yield additional improvements in trust-modelling accuracy with both PTM and TRAVOS, especially if the limiting step were such that it would yield a very small advisor network. We believe that these techniques will be very useful for trust researchers seeking to improve the accuracy of their own trust models, and to that end we explain how other researchers could apply these modifications themselves, in order to identify the optimal parameters for their usage. We discuss as well the value of our proposals for identifying an "optimal" size for a social network, and the use of referral systems, for researchers in other areas of artificial intelligence.

Navigation of a group of autonomous agents that are needed to maintain a formation is a challenging task which has not been studied much in especially 3-D terrains. This thesis presents a novel approach to collision free path finding of multiple agents preserving a predefined formation in a 3-D terrain. The proposed method could be used in many areas like navigation of semi-automated forces (SAF) at unit level in military simulations and non player characters (NPC) in computer games. The proposed path finding algorithm first computes an optimal path from an initial point to a target point after analyzing the 3-D terrain data from which it constructs a weighted graph. Then, it employs a real-time path finding algorithm specifically designed to realize the navigation of the group from one way point to the successive one on the optimal path generated at the previous stage, preserving the formation and avoiding collision both. A software was developed to test the methods discussed here.

Multiagent systems (MAS) are systems consisting of several autonomous entities, called agents, which interact with each other to either further their own interests (competition) or in pursuit of a joint goal (cooperation). In systems composed of multiple autonomous agents, negotiation is a key form of interaction that enables groups of agents to arrive at a mutual agreement regarding some belief, goal or plan. The aim of this thesis is to develop a multiagent system that simulates the negotiation process between parties about sharing of cost overrun in international construction projects. The developed tool can be used to understand how the risks and associated costs are shared between parties under different scenarios related with the risk allocation clauses in the contract, objectives of parties and level of knowledge about actual sources of cost overrun. MAS can be utilized by decision-makers to predict potential outcomes of a negotiation process.

The complexity of the path finding and navigation problem increases when multiple agents are involved and these agents have to maintain a predefined formation while moving on a 3-D terrain. In this thesis, a novel approach for multiresolution formation representation is proposed, that allows hierarchical formations of arbitrary depth to be defined using different referencing schemes. This formation representation approach is then utilized to find and realize a collision free optimal path from an initial location to a goal location on a 3-D terrain, while preserving the formation. The proposed metod first employs a terrain analysis technique that constructs a weighted search graph from height-map data. The graph is used by an off-line search algorithm to find the shortest path. The path is realized by an on-line planner, which guides the formation along the path while avoiding collisions and maintaining the formation. The methods proposed here are easily adaptable to several application areas, especially to real time strategy games and military simulations.

Psychological studies have shown that information exchange is a key component of effective teamwork. In addition to requesting information that they need for their tasks, members of effective teams often proactively forward information that they believe other teammates require to complete their tasks. We refer to this type of communication as proactive information exchange and the formalization and implementation of this is the subject of this thesis. The important question that we are trying to answer is: under normative conditions, what types of information needs can agent teammates extract from shared plans and how can they use these information needs to proactively forward information to teammates? In the following, we make two key claims about proactive information exchange: first, agents need to be aware of the information needs of their teammates and that these information needs can be inferred from shared plans; second, agents need to be able to model the beliefs of others in order to deliver this information efficiently. To demonstrate this, we have developed an algorithm named PIEX, which, for each agent on a team, reasonably approximates the information-needs of other team members, based on analysis of a shared team plan. This algorithm transforms a team plan into an individual plan by inserting coomunicative tasks in agents' individual plans to deliver information to those agents who need it. We will incorporate a previously developed architecture for multi-agent belief reasoning. In addition to this algorithm for proactive information exchange, we have developed a formal framework to both describe scenarios in which proactive information exchange takes place and to evaluate the quality of the communication events that agents running the PIEX algorithm generate. The contributions of this work are a formal and implemented algorithm for information exchange for maintaining a shared mental model and a framework for evaluating domains in which this type of information exchange is useful.

Many biological systems are known to accomplish complex tasks in a decentralized, robust, and scalable manner - characteristics that are desirable to the coordination of engineered systems as well. Inspired by nature, we produce coordination strategies for a network of heterogenous agents and in particular, we focus on intelligent collective systems. Bottlenose dolphins and African lions are examples of intelligent collective systems since they exhibit sophisticated social behaviors and effortlessly transition between functionalities. Through preferred associations, specialized roles, and self-organization, these systems forage prey, form alliances, and maintain sustainable group sizes. In this thesis, we take a three-phased approach to bioinspiration: in the first phase, we produce agent-based models of specific social behaviors observed in nature. The goal of these models is to capture the underlying biological phenomenon, yet remain simple so that the models are amenable to analysis. In the second phase, we produce bio-inspired algorithms that are based on the simple biological models produced in the first phase. Moreover, these algorithms are developed in the context of specific coordination tasks, e.g., the multi-agent foraging task. In the final phase of this work, we tailor these algorithms to produce coordination strategies that are ready to be deployed in target applications.

A scenario where multiple entities interact with a common environment to achieve individual and common goals either co-operatively or competitively can be classified as a Multi-Agent System. In this thesis, we concentrate on the situations where the agents exhibit selfish, competitive and strategic behaviour, giving rise to interesting game theoretic and optimization problems. From a computational point of view, the presence of multiple agents introduces strategic and temporal issues, apart from enhancing the difficulty of optimization. We study the following natural mathematical models of such multi-agent problems faced in practice: a) combinatorial optimization problems with multi-agent submodular cost functions, b) combinatorial auctions with partially public valuations and c) online vertex-weighted bipartite matching and single bid budgeted allocations. We provide approximation algorithms, online algorithms and hardness of approximation results for these problems.