290 likes | 400 Views
Semantic Knowledge Model and Architecture for Agents in Discrete Environments. Michal Laclav í k Ústav informatiky Slovenskej akadémie vied. Outline. Motivation State of the Art Objectives Methodology and Tools Agent Knowledge Model – Models, Methodology, Library Applications Conclusion.
E N D
Semantic Knowledge Model and Architecture for Agents in Discrete Environments Michal Laclavík Ústav informatiky Slovenskej akadémie vied
Outline • Motivation • State of the Art • Objectives • Methodology and Tools • Agent Knowledge Model – Models, Methodology, Library • Applications • Conclusion
Motivation and State of the Art • MAS ispowerful paradigm for distributed or heterogeneous systems • MAS need Knowledge Support and Semantics • MAS need Connection with Existing Commercial Standards • Agent Technology Roadmap: Current MAS Systems – lack of Internal Agent Knowledge Model, lack of interconnection with semantic web results (knowledge model representations) and commercial standards • Focus on Agents and Knowledge representation (Ontologies) • Knowledge Management and Experience Management as application domains
State of the Art - Agents • Agent Definition: An agent is a computer system capable of flexible autonomous action in a dynamic, unpredictable and open environment. (LUCK 2003) • MAS Standards: FIPA, MASIF • Related to agent communication, agent platforms • No standards for internal agent knowledge model with available implementations
State of the Art - Agents • Architectures: • Reactive Architecture • No specification of knowledge model, behavior of agent is based on implemented responses to environment states • Belief Desire Intention Architecture – BDI • Belief – represents knowledge model, available some implementations based on logic programming, not used in FIPA compliant MAS • Behavioral Architecture • FIPA compliant MAS are based on such architecture • No specification of Internal Agent Knowledge model – depend on agent designer and developer • JADE Agent System • Support for ontologies based on FIPA-SL (Similar to First Order Logic) • No Query engine • No Storage • No Inference
Ontologies Knowledge Representation OWL-DL compatible with Description Logic Query and Storage Engines available RDF, OWL, RDQL based Application domain Knowledge Management (KM) is the process through which organizations generate value from their intellectual and knowledge-based assets (Source: CIO Magazine) Experience Management is special kind of KM – based on “lessons learned” Reasoning Actions Pragmatics Knowledge Semantics Information Data Syntax Characters (Bergman, 2002, Experience Management) State of the Art – Ontologies, Knowledge
User requests Displaying results Agent 1 XML, XML-RPC, SOAP ACL KM Agent 3 FIPA ACL, KIF, FIPA-SL, FIPA-RDF IIOP, HTTP, SMTP Agent 2 KM FIPA ACL, RDF/OWL, RDQL Knowledge Storage Querying Problem Specification Graphical User Interface Multi Agent System External System Knowledge Model Knowledge Base Directory Facilitator
State of The Art Conclusion • Focus on software, intelligent and FIPA compliant agents • Providing better semantic infrastructure (ontologies, knowledge models) • Apply basic principles of software and knowledge engineering • Make stronger connection between MAS and existing commercial technologies
Used Tools and Software • Protégé Ontology Editor • Support for OWL ontology format • Can be used as modeling tool • JADE (Java Agent DEvelopment Framework) • Most developing MAS framework • Compliant with FIPA standards • Jena – Semantic Web Framework for Java • Support for OWL – best available OWL API • Support for RDQL model querying
Agent Knowledge Model Objective: Design of Agent Architecture using Ontology based Knowledge Model
Based on Events, Resources, Actions, Actors, Context Formally Described using Sets, Description Logic (compatible with OWL-DL), Graph Representation Actor Context updating function/algorithm (Actor Environment State) CAnew = fC(ea,CAold) Resources updating function/algorithm (result of fulfilled actor goals) RAnew = fR(CAnew,RAold) Agent Knowledge Model
Software Development Methodology Objective: Design of Software Development Methodology for creation of Agents with Ontology based Knowledge Model
Development Methodology (Knowledge Model) • Extending Model with Protégé Editor following CommonKADS models • Organizational or Environment Model • Task Model • Agent or Actor Model • Includes implementation of algorithms for context and resource updating • Results • Ontology developed in Protégé which can be exported in OWL format. • Concrete Algorithms for each actor (often algorithms are similar or same) which updates actors' context CAnew and resources RAnew.
Development Methodology (System Design) • UML Diagrams for concrete Application Domain • Use Case Diagram • for each agent • agent is taken as system boundaries • Sequence Diagram • Communication among agents • Class Diagram • Behaviors are described as methods
Agent Software Library Objectives: Design of Agent Architecture using Ontology based Knowledge Model Design & Development of Software Library for building Intelligent Agents with Ontology Knowledge Model with possibility to plug agents to existing commercial technologies
Support for OWL based Agent Knowledge Model Support for XML-RPC connection to receive event and send plain XML Support for agent communication using FIPA ACL with OWL and RDQL as content languages Support for Presentation of Ontological Knowledge (RDF/OWL => plain XML + XSL => HTML) JADE and Jena Integration Available on JADE official website to MAS community Agent Software Library
Support for Knowledge and Experience Management Objective: Design of Generic Ontology Model for Experience Management with extension to different application domains.
Extended Agent Memory Model Workflow Related WfInstance, WfActivity ActiveHint Sub class of resource Representation of Experience Employee Extension of Model for Experience Management
Presentation of Ontology based Knowledge Objective: Design and Development of user friendly Knowledge Presentation for MAS.
Ontology Tree Browse window Graph XSL Transformation RDF/OWL => Plain XML + XSL => HTML Infrastructure to receive plain XML using XML-RPC Presentation of Ontology based Knowledge
Applications Objective: Evaluation of Results on real pilot operation.
Title: Platform for Organizationally Mobile Public Employees Duration: Sep 2002- Dec 2004 Knowledge Management to support employees Workflow based Administration Processes To support Employee Mobility in organization Agent Architecture based on autonomous co-operating agents Interaction Layer Process Layer Pellucid 5FP IST Project Pellucid Architecture Pellucid Agents
CDG, Genoa, ItalyTraffic Light Management MMBG, Sanlucar, SpainProject Management SADESI, Seville, SpainTelephone Incidence Resolution Pellucid Applications
Work on new EMBET architecture Current state: User Assistant Agent in K-Wf Grid uses model presented in thesis. Algorithms presented in chapter 5 were reused with same improvements and modifications. Architecture is not Agent based but users of system are modeled as actors. Knowledge Model, its implementation and modified algorithms presented in thesis are used Title: Knowledge-based Workflow System for Grid Applications Objectives: To support workflow construction and execution with Knowledge Duration: Sep 2004 - Feb 2007 K-Wf Grid 6FP IST Project
Conclusion (1) • The most significant scientific achievements • Agent knowledge model • Applicable in any discrete environment where actors need to be modeled • Can be expressed by ontology, sets or description logic • Such model was found useful for: • Simple goal oriented agents • Knowledge Management Solution based on Agents (Pellucid) • Experience Management Solution non agent based (EMBET System) • Development Methodology • Speed up Knowledge based Agent development for concrete application domains
Conclusion (2) • The most significant development achievements • Agent Library • Support for OWL based Agent Knowledge Model • Support for XML-RPC connection to receive event and send plain XML • Support for Presentation of Ontological Knowledge (RDF/OWL => plain XML + XSL => HTML) • Support for agent communication using FIPA ACL with OWL and RDQL as content languages • JADE and Jena Integration • Available on JADE official website to MAS community (year 2005 – 577 downloads)
Thank you ! Michal Laclavik