Copyright Cesar Augusto 6599 HEUDIASYC, UTC, A M ulti-Agent A rchitecture for K nowledge A cquisition Cesar TACLA* Jean-Paul BARTHES {cesar.tacla, CNRS UMR 6599 HEUDYASIC Université de Technologie de Compiègne Compiègne, France *CEFET / CAPES Curitiba, Brazil
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, Introduction Problem Lost of K = Time + Spread information + Dynamic groups How to organize knowledge in Research and Development teams (R & D) >> reuse of gained experience Why R & D groups ? –serious constraints on time –top-down methods are difficulty to apply –activities are rich, complex and sometimes unexpected
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, KM Environment Approach –An environment that integrates humans and artificial agents –Cooperative construction of a distributed group memory –Guidelines Minimize effort from knowledge engineering Minimize effort from knowledge producers
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, KM Environment Requirements for the KM Environment 1.Nature of K: technical K + design K 2.KM env functions: document management, and aid to the externalization and internalization 3.Integration: transparent, integrated into the individuals day- to-day activities, bottom-up K acquisition, distributed (there is no central node or filter) 4.Different perspectives: each individual organizes K according to her preferences, distributed
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, KM Environment A KM Environment is –a distributed group memory –composed by individual memories, –where individuals work cooperatively
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, KM Environment Docs Capture Clusters of Docs Classified docs + clusters Personal memory ClassifyCluster Conceptual Models Construction of a personal memory
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, KM Environment Personal memory 1 Preserve Conceptual Models Sources Project memory Personal memory i Familiarize Validate Personal memory 2 Personal memory 3 Research Cooperative processes
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, KM Environment K item evolution graph: distributed graph KI 1 KI 2 KI 3 Derived from modifies KI 4 refers to KI 5 replaces A’s memoryB’s memoryC’s memory KI = Knowledge Item
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, MA architecture Built on the OMAS platform –Open Multi-Agent System Features –Cognitive agents –Two agent types: Service and Personal Assistant –Coterie Agents share the same local network Communication is in broadcast Contract-net / Simple protocol (request, answer, inform) SPA
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, MA architecture User scope components –Each user has a staff of agents –Staff agents run on the same computer –Service agents in the staff are dedicated to the assistant agent Organizer User A’s staff S PA Assistant Interface Capture operations+docs Save (operations, docs) Cluster Classify
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, MA architecture Repository Agent Transfer Agent GW or DB X S S Project Agent Project scope K preservation (preserve) Sources of the conceptual models (familiarize) Communication Doc retrieving/storing Access rights Components project level Agents run on different machines
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, MA architecture Repository Agent Transfer Agent GW or DB X A’s staff S PA S S Project Coterie (LAN) Project Agent B’s staff S PA Project scope organizer Proposed Architecture
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, MA architecture Desktop operations Fred’s staff S PA Organizer Save operations + docs Cluster Classify Assistant Capture operations + docs P S S S PA S S Fred Coterie members Reuse Retrieve Diffuse Validate
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, MA architecture Personal Memory A Personal Memory B Project Memory Derived from modify replace Distributed group memory Source of a conceptual model
Copyright Cesar Augusto 6599 HEUDIASYC, UTC, Conclusion Future work –Define the semantics for the relations in the evolution graph –Define a protocol for coordinating the validation process Important issues –Specification of the minimal components (functions and knowledge items) for a KM environment. –The notion of staff of agents dedicated to a user –Aid the user to formalize his/her Knowledge -> gradual formalization of information –Augmenting the knowledge sharing (expected !) –Decreasing the information load