Intelligent Software Agent Technology by kfl11257

VIEWS: 28 PAGES: 25

									Intelligent Software Agent
        Technology



                  Dr. Thomas E. Potok
        Collaborative Technologies Group Leader
       Computer Science and Mathematics Division
             Oak Ridge National Laboratory

                      Research Team
  Mark Elmore, Joel Reed, Nagiza Samatova, Jim Treadwell
Oak Ridge National Laboratory

            Established in 1943 for the World War II
             Manhattan Project.
            ORNL today pioneers the development
             of new energy sources, technologies,
             and materials and the advancement of
             knowledge in the biological, chemical,
             computational, engineering,
             environmental, physical, and social
             sciences.
            Budget: $870 million, 80% Department of
             Energy, 20% work for others.
            3800 employees, 1500 scientists and
             engineers




                                                    2
   Agent Technology Trend

                                                  Internet
                                      Telephone

                         Face to Face




“The Semantic Web… will open up the knowledge and workings of
humankind to meaningful analysis by software agents, providing a
new class of tools by which we can live, work and learn together”
Tim Berners-Lee, James Hendler, Ora Lassila
“The Semantic Web” Scientific American 5/2001


                                                                3
       Successful Projects
       We have extensive expertise in                      Scientific Data Management
       agent development                                VIPAR Knowledge Discovery
       Began working with agent                     Supply Chain Management Agent System
       technologies in 1980s                      Manufacturing Emulation Agent System
                                               Collaborative Management Environment
                                               Neural Nets for Recovery Boiler Control
                                        Neural Nets for Bankruptcy Prediction
                                   Neural Nets for Spring-back Prediction
                             Collaborative Design System
                         Neural Nets for Resistance. Spot Welding
                   Neural Nets for Material Mix Optimization
              Genetic Algorithms for Chemical Synthesis              Over 10 successful projects
          Knowledge-based Systems - Manufacturing Advisors           within the last 5 years
     Knowledge-based Systems for Constructability                    Collaborations with leading
   Design and Analysis of Computer Experiments
Knowledge-based Computer Systems Calibration
                                                                     agent experts

1985               1990                1995               2000


                                                                                              4
Collaborative Management Environment
              Managing heterogeneous information

Original Document                        XML tagged Document
                                     <!ELEMENT   c m e - p r o p o s a l ( p r o p o s a l -r e c - i t e m ) + >   <cme-proposal>
                                     <!ELEMENT proposal-rec-item                                                          <proposal-rec-item>
                                            (proposal |                                                                         <proposal>
                                            staff-resources |
                                            funds-resources |                                                                         <ben> 19980727115540408 </ben>
                                            equipment-resources |                                                                     <work-prop-no>
                                            budget-attachment |
                                            equipment-attachment |                                                                          ERKJE09
                                            major-cost-attachment |                                                                    </work-prop-no>
                                            time |
                                            milestone-task |
                                                                                                                                      <work-prop-title>
                                            doe-program |                                                                                   Collaborative Management Environment
                                            doe-program-manager |                                                                     </work-prop-title>
                                            oo-reviewer |
                                            pi |                                                                                      <rev-no>
                                            proposal-manager |                                                                              0
                                            hq-organization |
                                            operations-office |                                                                        </rev-no>
                                            attachment |                                                                              <inst-plan-flag>
                                            contractor |
                                            funds-actual |                                                                                  Y
                                            cost-actual |                                                                             </inst-plan-flag>
                                            report-requirements
                                            )+>                                                                                       <work-prop-desc>
                                                                                                                                            The Collaborative Management Environment (CME) is a joint
                                     <!ELEMENT proposal                                                                                     research project between the Oak Ridge National
                                            (ben         |
                                            work-prop-no |                                                                                  Laboratory (ORNL) and Lawrence Berkely National
                                            work-prop-title |                                                                               Laboratory (LBNL) that is funded by the U.S. Department
                                            rev-no |
                                            inst-plan-flag |                                                                                of Energy (DOE). CME is designed to support …
                                            work-prop-desc |                                                                          </work-prop-desc>
                                            fwp-rpt-rqmt
                                            )+>                                                                                       <fwp-rpt-rqmt>
                                                                                                                                            Results will be reported in periodic highlights to DOE
                                     <!ELEMENT ben (#PCDATA)>
                                     <!ELEMENT work-prop-no (#PCDATA)>                                                                      and in peer-reviewed journals and conference proceedings.
                                     <!ELEMENT work-prop-title (#PCDATA)>                                                             </fwp-rpt-rqmt>
                                     <!ELEMENT rev-no (#PCDATA)>
                                     <!ELEMENT inst-plan-flag (#PCDATA)>
                                                                                                                                </proposal>
                                     <!ELEMENT work-prop-desc (#PCDATA)>                                            …
                                     <!ELEMENT fwp-rpt-rqmt (#PCDATA)>                                                    </proposal-rec-item>
                                     …                                                                              </cme-proposal>




Field level query                        Financial Summary




                                                                                                                                                                                             5
Portfolio Management Environment

 Dr. Ernest Moniz, Under Secretary of
  Energy, approves
   Portfolio Management Environment (PME)
   Based on the ORNL Collaborative Management
    Environment (CME) project
 Estimated 5-year implementation cost
   $17 million total investment
   Producing approximately $39 million annual
    productivity gains for DOE


                                                 6
   SDM
   center    Agents in Scientific Data
             Management
                                   Probe
                                 Parallel   Distributed
                                 Agent      Clustering




Astrophysics                                                 Scientific
                       HPSS
Climate                                                      Analysis
Genomic                                                      And
Simulations                                                  Discovery

            DOE Labs: Argonne, Berkeley, Livermore, Oak Ridge
            Universities: Georgia Tech, NC State, Northwestern, SDSC

  3 year $9M Center to manage massive scientific data

                                                                          7
     What are Agents?
 ...Software entities that assist people and act on their behalf - IBM

Traditional Software                               Agents

 Client Server Messaging                  Peer to Peer Messaging
                                                   Reactive
        Behavior                                    Model
                                                  Knowledge
                                                    Model
          State
                                                    Social
                                                    Model




                                                                         8
     What are Agents?
 ...Software entities that assist people and act on their behalf - IBM

Traditional Software                               Agents

Client Server Messaging                   Peer to Peer Messaging
                                                   Reactive
        Behavior                                    Model
                                                  Knowledge
                                                    Model
          State
                                                    Social
                                                    Model




                                                                         9
       Simple Agent Example
        Agent,
   find me the book                                  2 Days
                            Dedicated    Amazon
   “War and Peace,”                                  $18.50
                             Agents
      and I need it
       tomorrow                         Barnes and   1 Day
                                          Nobel      $21.75


                                                     1 Day
                                        B. Dalton
                                                     $20.25




 Does the     Form a plan
   agent       to buy the
understand        book
  buying
  books?


                                                              10
       Simple Agent Example
        Agent,
   find me the book                                                   2 Days
                             Dedicated      Amazon
   “War and Peace,”                                                   $18.50
                              Agents
      and I need it
       tomorrow                           Barnes and                  1 Day
                                            Nobel                     $21.75


                                                                      1 Day
                                           B. Dalton
                                                                      $20.25

                              Agent
                            Communities
                                                                      1 Day
                                             Library
                                                                       Free
 Does the     Form a plan
   agent       to buy the
                                                                      1 Day
understand        book                          eBay
                                                                      $12.50
  buying
  books?                    Execute the                               Now
                                            On-Line
                               plan       www.online-literature.com   Free


                                                                               11
Use of Intelligent Agents at US
       Pacific Command
             VIPAR
    USPACOM Command Virtual Information Center
      • Only 10% of available newspapers scanned
      • Entirely manual operation
Manually gather    Manually Organize           Manually Analyze
Information from   Information                 And Summarize
20–30 newspapers                               Information
Inside China




Russia Today                                                CINC Summary




Jakarta Post
                                           Analysts
                                                            Summary
                   Unorganized Newspaper
                           Articles


                                                                    13
               New Intelligent Agent Approach
     • Capable of organizing 10,000 article per day
     • Requires only a desktop computer
                Automatically Gather,            Manually Analyze
                Organize, and Summarize          And Summarize
Inside China    Thousands of articles            Information

                                                               CINC Summary
Russia Today




Jakarta Post
                                                               Summary
                                              Analysts
                    Article    Organize and
                   Retrieval    summarize
                    Agents       Articles




                                                                        14
VIPAR Architecture Overview

                Information      Whiteboard Agent         Text Analysis
    Internet       Agents                                     Agent
                                                               D1:   D2:   D3:
                                                         T1:    1     0     0
                                                         T2:    1     0     1
                Information                              T3:    0     0     1
                                                         T4:    0     1     0
   Database        Agents                                T5:    1     1     1
                                                         T6:    0     1     1
                                                         T7:    1     0     1


                    …
                                                               Cluster
                Information          Agent                      Agent
    Agent
  Communities      Agents          Communities




  Multi-agent ORMAC Framework                       RDF Ontologies
  Java                Java RMI                      FIPA ACL


                                                                                 15
         Clustering: A Simple Example
1. Original and Parsed Documents                                            2. Terms Dictionary            3. Vector Space Model
Document Title                                    Key Terms                  T1:   Aircraft
                                                                                                             T1:
                                                                                                                     D1:
                                                                                                                      1
                                                                                                                           D2:
                                                                                                                            0
                                                                                                                                  D3:
                                                                                                                                   0
         China Intercepts U.S.                    China US Aircraft          T2:   China                     T2:      1     0      1
D1:                                                                          T3:   Expert                    T3:      0     0      1
         Military Aircraft                        Military                                                   T4:      0     1      0
                                                                             T4:   Fleet
         Russia to Extend Life of                 Russia Nuclear Military                                    T5:      1     1      1
D2:                                                                          T5:   Military                  T6:      0     1      1
         Nuclear Military Fleet                   Fleet                                                      T7:      1     0      1
                                                                             T6:   Russia
         U.S. -Based Chinese Military             US China Military          T7:   US
D3:
         Expert Denied Release                    Expert Release

     4. Document Similarity                                                        5. Euclidean Distance
                                                                                                     k d
                                ( i )( j )T                                          
      Sij                                                                         D e (υi , υ j )   (υik  υ j k ) 2
              ( i )( j )T  ( i   j )( i   j )T
                                                                                                    k 1


     6. Agglomerative Hierarchical Clustering                                      7. Phylips tree representation

                                                                                                                             D1

                                                                                     D2

                                                                                                                             D3
              1     2       3       4      8        6     7




                                                                                                                                   16
VIPAR “Nuclear Weapons” Search - 9/21/00
       How are these articles organized?
                      Wen Ho Lee Spends First Day Savoring Home
                       Delights
                      Lee Case Points up Scientists' Attitude
                       on Security
                      India and Pakistan: Troubled relations
                      U.S. China Trade Vote Milestone on Rocky
                       Road
                      Troops die in Kashmir clashes
                      Clinton Concerned Over Lee Case - Reno On
                       Defensive
                      IAEA Supports Putin Nuclear Power
                       Initiative
                      China Rejects Moves to Tighten Regulation
                       of Nuclear Materials
                      Clinton Calls For Review Of Lee Secrets
                       Case
                      Asian-Americans Demand White House Action
                       on Lee




                                                               17
Manual Organization
“Business as Usual”
   Wen Ho Lee
       1.   Wen Ho Lee Spends First Day Savoring Home Delights
       2.   Clinton Calls For Review Of Lee Secrets Case
       3.   Clinton Concerned Over Lee Case - Reno On Defensive
       4.   Asian-Americans Demand White House Action on Lee
       5.   Lee Case Points up Scientists' Attitude on Security
   India and Pakistan
       6.   India and Pakistan: Troubled relations
       7.   Troops die in Kashmir clashes
   IAEA Meeting
       8.   IAEA Supports Putin Nuclear Power Initiative
       9.   China Rejects Moves to Tighten Regulation of Nuclear Materials
   U.S. China Trade
       10. U.S. China Trade Vote Milestone on Rocky Road




                                                                             18
         VIPAR Automatic Organization
         Same as manual
             3       2                                                Wen Ho Lee
                                                                       1.    Wen Ho Lee Spends First Day Savoring Home
     4                         1                                             Delights

                      Wen Ho Lee                                       2.    Clinton Calls For Review Of Lee Secrets Case
5                                                                      3.    Clinton Concerned Over Lee Case - Reno On
                                                                             Defensive
                                                                       4.    Asian-Americans Demand White House
                                                                             Action on Lee
                                                                       5.    Lee Case Points up Scientists' Attitude on
                                                                             Security

                                                                      India and Pakistan
                                                                       6.    India and Pakistan: Troubled relations
                                                                       7.    Troops die in Kashmir clashes

    India/Pakistan                                US China Trade      IAEA Meeting
                                                                       8.    IAEA Supports Putin Nuclear Power Initiative
6
                                                                       9.    China Rejects Moves to Tighten Regulation of
                                   IAEA Meeting              10              Nuclear Materials

         7
                                                                      U.S. China Trade
                                                                       10.   U.S. China Trade Vote Milestone on Rocky
                                          9                                  Road
                           8



                                                                                                                          19
Endorsements

 US Pacific Command’s CINC Admiral Blair
   “A tremendously successful project”
   “Software agents … lead to substantially
    improved analytical products.”
 Science and Technology Advisor
   “a grand slam home run!”
   “first time we've seen information discovery
    and knowledge management software working
    at HQ USCINCPAC operationally.”


                                                   20
Enhanced VIPAR Performance


                 The Phylips Tree
                  representation of the
                  clustering of 1015
                  articles
                 Required almost 360
                  second (6 minutes)
                  and 241 MB of
                  memory to compute



                                          21
Battelle Funded VIPAR Upgrades

 Memory
   Evaluated caching of document title and
    collection
   Optimized internal data structures
   Experimented with clustering partial articles.
   Applied Porter’s Stemming Algorithm
 Clustering Performance
   Clustering algorithm “Clusters in place”
   Hybrid clustering approach to large article sets

                                                       22
Result
          The Phylips Tree
           representation of
           the clustering of
           9461 articles
          Required 2.5
           minutes (137
           seconds) and 495
           MB of memory to
           compute




                               23
Summary
 Intelligent agents are a significant future
  technology direction
 We have a significant progression of
  deployed agent systems and research to
  our credit
 We have ringing endorsements regarding
  the success of our systems and research
 We look forward to the next step…


                                                24
Contact Information
      Tom Potok
      PotokTE@ornl.gov
      865-574-0834

      Mark Elmore
      ElmoreMT@ornl.gov
      865-241-6372

      Jim Treadwell
      JZT@ornl.gov
      865-574-3350


                          25

								
To top