Generative Attention Networks for Multi-Agent Behavioral Modeling

Semantic Web supports a fire-new infrastructure for solving the problem of semantic information interoperability, and it promises to support an intelligent and automatic information-processing platform for multi-agent system whose ultimate objective is to provide better services for end-users, for example, interoperable information query. Therefore, except agent-to-agent interaction in multi-agent system, there is human-to-agent interaction. To unify the two kinds of interaction, this paper introduces Q language - a scenario description language for designing interaction among agents and users. A Q-based architecture, which integrates ontology servers, ontology-mapping servers, semantic information sources, and multi-agent query system, is presented as a system solution to semantic information interoperability on Semantic Web. Furthermore, we investigate key technologies of interoperability: domain ontology, ontology-mapping service, and related multi-agent system, and give an implementation to demonstrate how our architecture works.

Multi-agent systems (MAS) offer to solve problems in complex system and application domains, but are also complex environments. Standardization in the area of agent communication has led to development of MAS frameworks, simplifying MAS development. Agent interactions are essential for agents to exhibit their social abilities, and interaction scenarios for commonly known, standard interaction protocols, can be seen as building blocks of successful multi-agent systems. This article presents an introduction to interaction scenario modelling and code generation techniques, allowing quick prototyping of agent interactions and creation of agents implementing scenario roles

An approach of the resource conversion processes as a model for formalization of the processes occurring in organizational and technical systems is considered. For analysis and optimization of such processes, simulation is widely used. The disadvantage of using computer modeling is the need to formulate such an experiment plan with the model that would contain an optimal control impact. This is impossible without using of additional heuristic or experiments planning methods. Analysis of hybrid heuristic methods for optimizing of the resource conversion processes is presented. These methods integrate simulation modeling and artificial intelligence methods, in particular expert systems, multi-agent and evolutionary modeling. Integration of simulation and heuristic methods for solving the problem considered is confirmed. Application of genetic algorithm for optimizing the controlled parameters of the simulation model or expert system is considered. A hybrid method of using the expert system for solving the problem of optimizing the operation of genetic algorithm is considered. Application of evolutionary modeling for centralized and decentralized coordination of agent behavior in multiagent systems is also considered. A hybrid heuristic method for solving the problem of forming various types of the agent's interactions in the multi-agent systems depending on the relationships of agents to common resources has been analyzed. Choice of the hybrid method for integrating multi-agent simulation and genetic algorithm for solving the problem of optimizing resource conversion processes is substantiated. This method is implemented in the multi-agent genetic optimization information technology in the BPsim. MAS dynamic situations modeling system and in the metallurgical enterprise information automated system.

Interaction is a subject widely investigated in multi-agent systems (MASs), but some issues are still open. While most of current approaches of interaction in MAS just consider the interaction between agents, some problems are better modeled when the MAS is composed of agents, environment, interaction, and organization. In our approach, we integrate the interaction with the other MAS components, like the organization and the environment, keeping it as a first class abstraction. In this paper we present a conceptual model for the interaction component, a programming language to specify the interaction, and how our approach was integrated in an MAS platform. The main result of this paper is the conception of the interaction as a first class abstraction considering an MAS composed of agents, environment, interaction, and organization.

Development of agent systems is naturally a complex task due to the fundamental characteristics of agents. In addition, agent internals and inter-agent behavior models inside Multi-agent Systems (MAS) may become even more difficult to implement when interactions of agents with web services on the Semantic Web are taken into account. Our approach consists of the utilization of a Domain-specific Modeling Language (DSML) during MAS development in order to cope with the abovementioned challenge. This paper describes how the formal semantics of this DSML can be defined by especially focusing on its viewpoint on agent-semantic service interactions and discusses the use of this semantics definition on MAS validation. Determined semantic rules are both defined and implemented by using Alloy specification language which has a strong description capability based on both relational and first-order logic.

One of the most critical issues in the engineering of multi-agent systems (MAS) is the inadequacy of the available tools for MAS development and deployment. As we assume interaction as a first-class issue in MAS, tools are particularly required to monitor and debug inter-agents aspects, such as interaction protocols, coordination policies, social norms, and environment constraints. Since we claim that the definition of such tools is a basic research issue, in this paper we aim to identify the main requirements for development and deployment tools within an effective agent infrastructure. Focusing on agent interaction aspects, we take tuple-based coordination infrastructures--in particular the TuCSoN technology and tools--as our reference, and discuss the role of tools in a simple case study: the development and deployment of a well-known agent interaction protocol, the Contract Net.

Recent research on Distributed Artificial Intelligence (DAI) has focused upon agents’ interaction in Multiagent Systems. This paper presents a text understanding oriented multiagent dynamic interaction testbed (TUMIT): the theoretic framework based upon game theory, the free-market-like system architecture, and experimentation on TUMIT. Unlike other DAI testbeds, TUMIT views different text understanding (TU) methods as different “computational resources”, and makes agents choose different TU paths and computational resources according to the resource information on the bulletins in their hostcomputer. Therefore, in TUMIT, task allocation is wholly distributed. This makes TUMIT work like a “free market”. In such a system, agents’ choices and resource load may oscillate. It is shown theoretically and experimentally that if agents use multi-level of “history information”, their behavior will tend to converge to a Nash equilibrium situation; and that if agents use “recall-forget” strategy on “history information”, the convergence can be accelerated and the agents can acclimate themselves to changed environment. Compared with other DAI testbeds, TUMIT is more distributed, and the agents in TUMIT are more adaptive to the dynamic environment.

Existing middleware platforms for multi-agent systems (MAS) do not provide general support for observation. On the other hand, observation is considered to be an important mechanism needed for realizing effective and efficient coordination of agents. This paper describes a framework called Agent Observable Environment (AOE) for observation-based interaction in MAS. The framework provides 1) possibility to model MAS components with RDF-based observable soft-bodies, 2) support for both query and publish/subscribe style ontology-driven observation, and 3) ability to restrict the visibility of observable information using observation rules. Additionally, we report on an implementation of the framework for the JADE middleware platform, where AOE is realized as a custom kernel service.

Multi-agent systems are proposed as a solution to mitigate nowadays software requirements: open and distributed architectures with dynamic and adaptive behaviour. Like any other software, multi-agent systems development process is error-prone; thus testing is a key activity to ensure the quality of the developed product. This paper sheds light on agent testing as it is the primary artefact for any multi-agent system’s testing process. A framework called JADE Testing Framework (JTF) for JADE platform’s agent testing is proposed. JTF allows testing agents at two levels: unit (inner-components) and agent (agent interactions) levels. JTF is the result of the integration of two testing solutions: JAT a well-known framework for JADE’s agent’s interaction testing and UJade, a new solution that was developed for agent’s unit testing. UJade provides also a toolbox that allows for enhancing JAT capabilities. The evidence of JTF usability and effectiveness in JADE agent testing was supported by an empirical study conducted on seven multi-agent systems. The results of the study show that: when an agent’s code can be tested either at agent or unit levels UJade is less test’s effort consuming than JAT; JTF provides better testing capabilities and the developed tests are more effective than those developed using UJade or JAT alone.

Trace expressions are a compact and expressive formalism, initially devised for runtime verification of agent interactions in multiagent systems, which has been successfully employed to model real protocols, and to generate monitors for mainstream multiagent system platforms, and generalized to support runtime verification of different kinds of properties and systems. In this paper we formally compare the expressive power of trace expressions with the Linear Temporal Logic LTL, a formalism widely adopted in runtime verification. We show that any LTL formula can be translated into a trace expression which is equivalent from the point of view of runtime verification. Since trace expressions are able to express and verify sets of traces that are not context-free, we can derive that in the context of runtime verification trace expressions are more expressive than LTL.

A crucial part of multi-agent system design is the design of agent interactions. Traditional approaches to designing agent interaction use interaction protocols, which focus on defining legal sequences of messages. Such approaches do not naturally exhibit flexibility and robustness, and are not a good match for intelligent software agents which are autonomous, proactive, flexible and robust. The Hermes approach to designing agent interaction uses interaction goals, actions, and a number of failure recovery mechanisms to give a design methodology which is a good fit with intelligent software agents. However, the Hermes approach only covers part of the design process. In this paper we integrate Hermes with the Prometheus methodology, thus providing a complete methodology for designing multi-agent systems where interaction design is goal-oriented, yielding flexible and robust interactions.

This work presents a cyclic pursuit model based on non-linear inter-agent interactions driven by double-integrator dynamics. Cyclic pursuit models, beginning with the famous “Three-Bug problem”, have spurred a strong interest in the research community, encompassing linear and non-linear control models, synchronous and asynchronous pursuit and other similar variants and extensions. Much of the work, however, has been towards the development of kinematic pursuit models. Double-integrator models with directed interactions such as in cyclic pursuit models present strong challenges in the evaluation of the system stability and the emergence of global dynamic attributes based on local non-linear agent interactions. This work proposes and evaluates a control model based on a particular form of agent interactions involving linear attractive and non-linear repulsive forcing functions, directed from each agent to its leading agent along the pursuit curve. Specifically, the non-linear controller enforces a stricter implementation of collision avoidance among agents during pursuit, allowing for a reversal in pursuit direction when inter-agent separations drop to low values. Starting from arbitrary initial conditions for the multi-agent system, the emergence of stable cyclic pursuit configurations purely from local agent interactions is demonstrated. The proposed work has strong potential in co-operative perimeter-tracking applications such as wildfire monitoring and border patrol, where the need for efficient spatial distribution of agents around perimeters while in pursuit is essential for efficient data gathering and developing improved situational awareness for dynamic events.Copyright © 2012 by ASME

Verifying that agent interactions in a multiagent system (MAS) are compliant to a given global protocol is of paramount importance for most systems, and is mandatory for safety-critical applications. Runtime verification requires a proper formalism to express such a protocol, a possibly non intrusive mechanism for capturing agent interactions, and a method for verifying that captured interactions are compliant to the global protocol. Projecting the global protocol onto agents’ subsets can improve efficiency and fault tolerance by allowing the distribution of the verification mechanism. Since many real MASs are based on JADE, a well known open source platform for MAS development, we implemented a monitor agent that achieves all the goals above using the “Attribute Global Types” (AGT) formalism for representing protocols. Using our JADE monitor we were able to verify FYPA, an extremely complex industrialMAS currently used by Ansaldo STS for allocating platforms and tracks to trains inside Italian stations, besides the Alternating Bit and the Iterated Contract Net protocols which are well known in the distributed systems and MAS communities. Depending on the monitored MAS, the performances of our monitor are either comparable or slightly worse than those of the JADE Sniffer because of the logging of the verification activities. Reducing the log files dimension, re-implementing the monitor in a way independent from the JADE Sniffer, and heavily exploiting projections are the three directions we are pursuing for improving the monitor’s performances, still keeping all its features.

Research has recently grown on multi-agent systems (MAS) and their coordination and secure cooperative control, for example in the field of edge-cloud computing. MAS offers robustness and flexibility compared to centralized systems by distributing control across decentralized agents, allowing the system to adapt and scale without overhaul. The collective behavior emerging from agent interactions can solve complex tasks beyond individual capabilities. However, controlling high-order nonlinear MAS with unknown dynamics raises challenges. This paper proposes an enhanced genetic algorithm strategy to enhance secure cooperative control performance. An efficient encoding method, adaptive decoding schemes, and heuristic initialization are introduced. These innovations enable compelling exploration of the solution space and accelerate convergence. Individual enhancement via load balancing, communication avoidance, and iterative refinement intensifies local search. Simulations demonstrate superior performance over conventional algorithms for complex control problems with uncertainty. The proposed method promises robust, efficient, and consistent solutions by adapting to find optimal points and exploiting promising areas in the space. This has implications for securely controlling real-world MAS across domains like robotics, power systems, and autonomous vehicles.

The coalition formation process, in which a number of independent, autonomous agents come together to act as a collective, is an important form of interaction in multiagent systems. When effective, such coalitions can improve the performance of the individual agents and/or of the system as a whole. However, one of the main problems that hinders the wide spread adoption of coalition formation technologies is the computational complexity of coalition structure generation. That is, once a group of agents has been identified, how can it be partitioned in order tomaximise the social payoff? This problem has been shown to be NP-hard and even finding a sub-optimal solution requires searching an exponential number of solutions. Against this background, this paper reports on a novel anytime algorithm for coalition structure generation that produces solutions that are within a finite bound from the optimal. Our algorithm is benchmarked against Sandholm et al.ýs algorithm [8] (the only other known algorithm for this task that can also establish a worst-case bound from the optimal) and is shown to be up to 10^379 times faster (for systems containing 1000 agents) when small bounds from the optimal are desirable.