The Knowledge Grid

1. The Knowledge Grid Adam Belloum Computer Architecture & Parallel Systems group University of Amsterdam adam@science.uva.nl

2. Acronyms • KB  ???? • KF  Knowledge Fusion • KL  Knowledge Logistics • KM  Knowledge Management • KS  Knowledge Sources • KSNet-approach  Knowledge Source Network

3. Introduction • Knowledge Management is defined as a complexset of relations between people, processes, and technology bound together with the culturalnorms • like mentoring and knowledge sharing, which constitute an organization’s social capital. • Knowledge Management includes the following major tasks: • knowledge discovery • knowledge entry, capture of tacit knowledge, KF, etc., • knowledge representation • KB development, knowledge sharing and reuse, knowledge exchange, etc. • knowledge mapping • identifying KSs, indexing knowledge, making knowledge accessible.

4. Different approaches to KM • Different approaches proposed for solving KM based on • Algorithms of data search/retrieval in large databases, • Technologies of data storage and representation, etc. • KM research projects and tools: • Knowledge searching/retrieving from different types of documents: • Microsoft Share-Point Portal , SearchServer/KnowledgeServer , Text-To-Onto • Knowledge acquisition from experts and tacit knowledge revealing • Disciple-RKF, EXPECT, COGITO, OntoKick • Ontologies engineering • OntoEdit, Protégé, Ontolingua • KBs organization and development • HPKB • knowledge and information integration • KRAFT, InfoSleuth, RICA, OBSERVER

5. Possible Application domains of KM • Large-scale dynamic systems • Distributed operations in an uncertain and rapidly changing environment, where the information collection, assimilation, integration, interpretation, and dissemination are needed. • Focused logistics operations and/or web-enhanced logistics operations addressing sustainment, transportation, and end-to-end rapid supply to the final destination. • Distributed information management and real-time information/knowledgefusion to support continuous information and knowledgeintegration and exchange between all participants of the operations are needed. • Markets via partnerships with different organizations, where the dynamic identification and analysis of information sources and providing for interoperability between market participants in a semantic manner are needed.

6. KL is based • on individual user requirements, • available KSs, • and content analysis in the information grid environment. • Systems performing KL must • react dynamically to unforeseen changes and unexpected user needs, • keep up-to-date resource value assessment data, • support rapid execution of complex operations, • and deliver personalizedresults to the users/knowledge customers. • The proposed approach to KL is based on the KF technology and therefore assumes integration of knowledge from • different sources (heterogeneous) into a combined resource • to complement insufficient knowledge and obtain new knowledge.

7. Knowledge Source Network • A network of loosely coupled sources located in the information grid is known as • “Knowledge Source Network” (KS network). • The term “KS network” originates from the concept of VO based on the synergistic use of knowledge from multiple sources.

8. Knowledge logistic and system structure • Repositories have different structures, architectures and implementations depending on a purpose of a KMarea the repository belongs to. • Three components in the repository structure are defined: • a semantic component is used for knowledge representation in a common notation and terms; • a service component is used for knowledge indexing and search; • a physical component is used for knowledge storage and reuse.

9. The Semantic Component • An ontology is a formaldescription of entities and their properties, relationships, constraints, and behaviors. • It provides a commonterminology that captures key distinctions and is generic acrossmanydomains, facilitating translation of concepts among these domains. • Contains ontologies used to describe the domainterms and correspondence between terms of different ontologies.

10. The Service Components • A knowledge map • including information about locations of KSnetwork units and information about alternativesources containing similar information and KSs. • Monitoringtools perform permanent checking of KSs availability and perform appropriate changes in the knowledge map. • The knowledge map facilitates and speedsup the process of the KSs choice. • A user profile • including an organized storage of information about a user, his (her) requests history, etc. • This element is used for a number of purposes • faster search due to analyzing and utilizing request history and user preferences, Just-before-Time request processing, etc..

11. The Physical Component • Contains internal KB used or storage and verification of knowledge: • entered by experts, • learnt from users (knowledge consumers), • obtained as a result of the KF process, • acquired from KSs which are not free, not easily accessible, etc.

12. Knowledge Representation Formalism • A formalism of object-oriented constraint networks has been chosen for the ontology representation. • An abstract KS network model is based on this formalism. • This solution was mainly motivated by such factors as • support of declarative representation, • efficiency of dynamic constraint satisfaction, • problem modeling capability, maintainability, reusability, and extensibility of the object-oriented technology.

13. Knowledge Representation Formalism • an ontology (A) is defined as A = (O, Q, D, C): • Ois a set of object classes (“classes”). • Q is a set of class attributes (“attributes”). • Dis a set of attribute domains (“domains”). • Cis a set of constraints.

14. Knowledge Representation Formalism C = CI ∪ CII ∪ CIII ∪ CIV ∪ CV ∪ CVI • CI = {cI }, cI = (o, q), o ∈ O, q ∈ Q • belonging of attributes to classes; • CII = {cII}, cII = (o, q, d), o ∈ O, q ∈ Q, d ∈ D • belonging of domains to attributes; • CIII = {cIII}, cIII = ({o}, True ∨ False), |{o}| ≥ 2, o ∈ O • classes compatibility (compatibility structural constraints); • CIV = {cIV}, cIV = o, o, type, o’ ∈ O, o ∈ O, o = o” • hierarchicalrelationships (structural constraints) “is a” defining class taxonomy (type = 0), and “has part”/“part of” defining class hierarchy (type = 1); • CV = {cV }, cV = ({o}), |{o}| ≥ 2, o ∈ O • associative relationships (“one-level” structural constraints); • CVI = {cVI}, cVI = f({o}, {q}) → True ∨ False, |{o}| ≥ 0, |{q}| ≥ 0, o ∈ O, q ∈ Q • functional constraints referring to the names of classes and attributes.

15. Ontology types • Top-level ontology • describes notation for ontology representation in the system; • Application ontology (AO) • describes an application domain in terms of domain and tasks & methods • PreliminaryKS ontology • describes KS in KS’s terms and the top-level ontology notation; • KS ontology (KSO) • contains correspondence between terms of KS and AO; • Preliminary request ontology • describes userrequest in user’s terms and the top-level ontology notation, • Request ontology (RO) • contains correspondence between terms of preliminaryrequest ontology and AO; • Domain ontology • represents static knowledge about a particular domain in terms of the domain • Tasks & methods ontology • Describes problem-solving knowledge in terms of a domain or terms general for several domains.

16. Relationships between ontology and Ontology Libraries

17. Ontology-driven methodology for knowledge logistics • User profiles are used during interactions to provide for an efficient personalized service. Every user request consists of two parts: • structural constituent containing the request terms and relations between them • parametric constituent containing additionaluser-defined constraints.

18. Muti-agents Architecture • Multi-agent systems offer an efficient way • to understand, manage, and use the distributed, large-scale, dynamic, open, and heterogeneous computing and information systems. • The Foundation for Intelligent Physical Agents (FIPA) Reference Model was chosen as a technological basis for Multi-agent systems • Provides standards for heterogeneousinteractingagents/agent-based systems • Specifies ontologies and negotiation protocols. • FIPA-base technological kernel agents used in the system are: • wrapper (interaction with KSs), • facilitator (“yellow pages” directory service for the agents), • mediator (task execution control), • user agent (interaction with users).

19. Muti-agents Architecture • Translation agent  terms translation between different vocabularies • KF agent  KF operation performance • configuration agent  efficient use of KSNet • ontology management agent  ontology operations performance • expert assistant agent  interaction with experts • monitoring agent  KSsverifications

20. Muti-agents Architecture • In order to increase rapidity of the KF process in the system “KSNet” the following supporting tasks were defined • the knowledge map creation utilizing alternative KSs ranking • KS network configuration based on the task of efficient KSs choice • user request processing based on constraint network processing.

21. Knowledge Fusion Patterns • In the system “KSNet” the knowledge fusion takes place during execution of a number of tasks. • Selectivefusion(AO and KSO creation). • New KS is created which contains required parts of the initial KSs. Initial KSs preserve their internal structures and autonomy. • Simple fusion(OL creation and maintenance). • New KS is created which contains initial KSs. Initial KSs preserve their internal structures and lose (partially or completely) their autonomy.

22. Knowledge Fusion Patterns • Extension(knowledge map and internal KB maintenance). • One of initial KS is extended so that it includes the required part of other initial KS which preserves its internal structure and autonomy. • Absorption(new KS connection to the system). • One of initial KSs is extended so that it includes other initial KS which preserves its internal structure and loses (partially or completely) its autonomy. • Flat fusion(KF for user request processing). • New KS is created which contains initial KSs. The initial KSs dissolve within new KS and do not preserve their internal structures and autonomy.

23. Knowledge Fusion Patterns • Two initial KSs (A and B) with some structures of primaryknowledge units are given. • A tacit relationship between two primary knowledge units, namely • a3 from A and • b2 from B. • It is necessary to fuse two sources preserving the internalknowledgestructure and revealing the above tacit relationship. • Use of the KF patterns accelerates the KF process due to typification of fusion schemes.

24. Genetic Algorithms for knowledge source network configuration • The task of efficient KSs choice can be defined as • a configuration of feasible (in accordance with a set of structural constraints) and efficient (in accordance with a given criteria) KS network • definition of a set of rules prescribing when to use a certain KS. • The system “KSNet” includes: • The AO containing some ontology (knowledge) elements • (OE) ({aj}): AO = (O, Q, D, C) = {aj}n j=1, • Where : nis the number of the OEs. • KSs Si containing some OEs and described by preliminary KSOs at a time instant t: • A(Sit) = (O(Sit), Q(Sit), D(Sit), C(Sit)) = {slit}, l = 1, . . . , Lit; i = 1, . . . ,m; t = 1, . . ., T, • Where : Litis the number of OEs of KSi, mthe number of KSs in the system, and Tthe lifetime of the system “KSNet”. • Knowledge map associating OEs of AO with KSs at a time instant t. • Such association is denoted by “→”, • statement “OE aj is associated with KS Sit” is denoted by (aj → Sit) : KMt = {(aj → Sit)}, aj ∈ A.KSO is an association of KS to AO : A(Sit) = {(aj → slit)}.

25. Genetic Algorithms for knowledge source network configuration • user requestR is received by the system it is decomposed into a set of sub-requests rk, to be associated with AO’s OEs (i.e. translated into the system’s terms). • This association is contained in the request ontology A(R). • The system “KSNet” has the request translated/decomposed into sub-requests associated with AO’s OEs. Such requests are denoted by • R : R = {rk}, k = 1, . . .,K, A(R) = {(rk → aj)}, rk ∈ R, aj ∈ A, • R = {aj}, ∀aj∃(rk → aj) ∈ A(R), Where K is the number of subrequests. • When the operations are completed a set of feasible decisions of the task DecR can be written as: • DecR = {decR} , decR = {(aj → Sit)}, aj ∈ A(R). • Reliability, costs or time required for request processing can be used as criteria of the decision: • Reli = fReli(decR), Cost = fCost(decR) and Time = fTime(decR). • A decision is considered effective (denoted by deceff R ) if the value of the goal function is minimal with the above constraints being true: • deceff R ∈ DecR∀decR ∈ DecR, fCost(deceffR ) ≤ fCost (decR).

26. Genetic Algorithms for knowledge source network configuration • Initially a random set of solutions is generated • The set is sorted according to the chosen criteria • Mutation mechanism is applied to the best solution to generate new solutions. • The newly generated solutions are sorted, and the new iteration is performed. • The process is stopped after a predefined number of iterations. • For GA application the following notations are used. • A feasible static decision decR is defined as the “chromosome” with the following structure: decR ={decR j,i}, where each decR j,i is a Boolean variable equal to 1 if KSi is used for obtaining OEk or to 0 otherwise. • To simplify further discussion the OEs of R are renumbered in the following way: R = {aj}, j = 1, . . . , ˜n, ˜n = |R|. Hence, decR represents a binary

27. Genetic Algorithms for knowledge source network configuration • Experiments with a basic GA for tasks of different dimensions have been performed, with KS’ parameters and knowledge maps being randomly generated. • The results indicate that the number of requiredcalculations for obtaining a quasi-efficientdecision even using the basic non-optimized GA is smaller than that in the exhaustivesearch method. Fig represents the ratio of calculations number for the exhaustive search method that for the GA, and this improvement grows non-linearly along with the task dimension growth.

28. References • Alexander Smirnov et al. “ Knowledge logistics in information grid environment” Future Generation Computer Systems 20 (2004) 61–79