Vladimir Samoylov

P2P Agent Platform: Implementation and Testing

Peer-to-Peer (P2P) computing, a novel paradigm for distributed information technology, is currently receiving ever increasing interest from both academia and industry. Recent efforts undertaken to integrate multi-agent and P2P architectures are one of such very promising new opportunities. Indeed, multi-agent system (MAS) may be thought of as a multitude of autonomous entities, and, therefore, structuring the agents as a P2P network of nodes may result in an architecture providing a new dimension for design of open MAS composed of a highly transient population of agents. This paper presents an implementation of a P2P Agent Platform providing transparent interaction for distributed P2P agents. The developed P2P Agent Platform implements the basic mandatory components assumed by the functional architecture proposed by the FIPA Nomadic Agents Working Group. This implementation is supported with a search mechanism, a function of an underlying P2P infrastructure. The platform verification is done via prototyping a P2P ground object detection MAS in which the agents situated on top of the distributed instances of the P2P Agent platform provide classification services.

Multi-agent Peer-to-Peer Intrusion Detection

Ever increasing use of heterogeneous networks including mobile devices and ad-hoc sensor networks signifies the role of such information systems’ properties as openness, autonomy, cooperation, coordination, etc. Agent-based service-oriented Peer-to-Peer (P2P) architecture provides attractive (if not unique) design and implementation paradigm for such systems. This trend implies coherent evolution of security systems, that put in use the notions of distributed security policy, distributed intrusion detection systems, etc.1, requiring novel ideas. The paper proposes new architecture for such security systems. This architecture provides cooperative performance of distributed security means (agents) supported by distributed meta-knowledge base implemented as an overlay network of instances of P2P agent platform set up on top of P2P networking provider. The paper also analyzes new issues of P2P security systems with the main emphasis on P2P training of security agents to correlation of alerts produced by other relevant agents. An artificially built case study is used to highlight the essence of P2P security agent training to P2P decision combining and to exhibit new problems.

Ontology-Based Context-Dependent Personalization Technology

Personalization, a topmost concern of modern recommendation systems (RS), is intended to predict individual motivation of a customer for this or that choice. It depends on many factors forming explicit and implicit decision context. The paper proposes RS personalization technology that focuses on ontology–based extraction of semantically interpretable context of each particular customer’s decisions from his/her historical data sample with the subsequent machine learning–based extraction of customer–centered feature set and personal cause–consequence decision rules. The technology is fully implemented by Practical Reasoning, Inc. and validated via several case studies.

Interaction of Agents and Data Mining in Ubiquitous Environment

The paper considers agent and data mining interaction in ubiquitous environment and analyses some specific, for such environment, knowledge discovery tasks. It focuses on agent-based applications organized as overlay network of autonomous agents set up on top of ubiquitous environment composed overlay network of FIPA compliant P2P agent platform instances set up on top of the P2P communication service providers. The paper presents a generic P2P learning algorithm intended to provide for re-configurability of the above mentioned overlay networks and experimentally demonstrates it using a case study, P2P intrusion detection system.

Modeling dynamic engineering design processes in PSI

One way to make engineering design effective and efficient is to make its processes flexible – i.e. self-adjusting, self-configuring, and self-optimizing at run time. This paper presents the descriptive part of the Dynamic Engineering Design Process (DEDP) modeling framework developed in the PSI project. The project aims to build a software tool to assist managers to analyze and enhance the productivity of the DEDPs through process simulations. The framework incorporates the models of teams and actors, tasks and activities as well as design artifacts as the major interrelated parts. DEDPs are modeled as weakly defined flows of tasks and atomic activities which may only “become apparent” at run time because of several presented dynamic factors. The processes are self-formed through the mechanisms of collaboration in the dynamic team of actors. These mechanisms are based on several types of contracting negotiations. DEDP productivity is assessed by the Units of Welfare collected by the multi-agent system which models the design team. The models of the framework are formalized in the family of DEDP ontologies.

Support for Analysis, Design, and Implementation Stages with MASDK

In spite of much research and development on agent-oriented software engineering methodologies and supporting software tools, the problem remains of topmost importance. Many efforts are still needed to make such methodologies and software tools practically applicable at an industrial scale. This paper proposes extension of the Gaia methodology with a formal specification language, making it possible to implement Gaia as a model-driven engineering process supported by a corresponding agent-based software development environment, MASDK 4.0. The paper outlines MASDK 4.0 through the extended Gaia, and demonstrates the technology supported by MASDK 4.0 on the basis of a fragment of a case study on autonomous air traffic control.

Transportation Logistics Services from Cloud

One of the remarkable recent business trends is shifting to cloud as more services as possible. This is peculiar for not only simple services like e-commerce, but also sophisticated ones making available, for the customers, to exploit SaaS (Software as Services) market model. The paper presents the developed abstract architecture of the SaaS transportation planning service. The core functionality of this architecture is generic model and associated protocol of auction -- based multi-agent resource allocation and scheduling procedure comprising standard phases of the auction managed by multiple generic policies. The latter is introduced as a standard object class that provides for flexibility of planning as well as reusability of the core planning software that is à critical property of the software implementing any web service.

Agent-based distributed decision-making in dynamic operational environments

The paper presents an approach and the supporting software tools for integration of intelligent multi-agent systems and the peer-to-peer (P2P) technologies to support the development of heterogeneous distributed decision making applications operating in dynamic ubiquitous environment. Particularly, the proposed technology integrates two software tools, MASDK 4.0 software tool and a P2P agent platform. MASDK 4.0 is provided with a graphical application specification language that supports the thorough and consistent conceptual analysis, detailed design, code generation and deployment of multi-agent systems applications including those operating in dynamic environments. P2P agent platform is a FIPA compliant middleware enabling transparent P2P interaction of agents operating in heterogeneous and ubiquitous environment. The platform is fully compatible with the multi-agent applications developed by MASDK 4.0. The combined application of MASDK 4.0 and P2P agent platform is demonstrated by a case study of the air traffic control and emergency management. In this case study a model of autonomous air traffic control is implemented by a set of distributed intelligent agents. This application is a typical example of a modem agent-based distributed decision making system operating in a dynamic P2P environment.

Automated P2P Learning in Agent-Based Classification Networks

Peer-to-Peer (P2P) computing is a novel computing paradigm receiving ever increasing attention of the research community. It provides new opportunities in design and implementation of large scale intelligent systems satisfying modern requirements to scalability, autonomy, mobility, fault tolerance, etc. This paradigm becomes particularly attractive if it is integrated with multi-agent systems. The paper is focused on P2P cooperative decision making and P2P machine learning of cooperation of autonomous agents in open P2P networks. It proposes P2P mechanism for decision combining utilized by autonomous decision making agents and for P2P learning of decision combining. The paper results are validated using case study, P2P agent-based intrusion detection system.

Agent-Based Customer Profile Learning in 3G Recommender Systems: Ontology-Driven Multi-source Cross-Domain Case

Advanced recommender systems of the third generation (3G) emphasize employment of semantically clear models of customer cross-domain profile learned using all available data sources. The paper focuses on conceptual level of ontology-based formal model of the customer profile built in actionable form. Learning of cross-domain customer profile as well as its use in recommendation scenario requires solving a number of novel problems, e.g. information fusion and data source privacy preservation, among others. The paper proposes an ontology-driven personalized customer profile model and outlines an agent-based architecture supporting implementation of interaction-intensive agent collaboration in two variants of target decision making procedure that are content-based and collaborative filtering both exploiting semantic similarity measures.