New Trends in Intelligent Systems

1
New Trends in Intelligent Systems

• Dr. Jay Liebowitz
• Professor
• Johns Hopkins University
• Jliebow1_at_jhu.edu

2
AI Past, Present, and Future, AI Magazine,
25th Anniversary Issue of AAAI, Vol. 26, No. 4,
Winter 2005

• We are a scientific society devoted to the study
  of artificial intelligenceAllen Newell, The
  First AAAI Presidents Message, 1980
• As AI matures, its focus is shifting from
  inward-looking to outward-looking. Some of the
  new concerns of the field are social awareness,
  networking, cross-disciplinarity, globalization,
  and open accessAlan Mackworth, Current AAAI
  President, July 2005

3
The Next 50 Years

• The Semantic Web is to KR as the Web is to
  hypertextJames Hendler, U. of Maryland
• AI has not yet succeeded in its most fundamental
  ambitions. Our systems are fragile when outside
  their carefully circumscribed domainsRod
  Brooks, MIT
• Reasoning programs still exhibit little common
  sensePatrick Winston, MIT

4
More Quotes

• Integrative research will be particularly
  challenging for research students. To do it,
  they must master a wide range of formal
  techniques and understand not just the
  mathematical details but also their place in
  overall accounts of intelligent behaviorHaym
  Hirsh, Rutgers University
• Another reason for the slow progress is the
  fragmentation of AIAaron Sloman, U. of
  Birmingham

5
Innovation, 2004 (Patent Applications
Filed)Financial Times, June 8, 2005, Thomson
Scientific
6
Patents Filed by Sector in 2004 (Spain)
Financial Times, Oct. 26, 2005, Thomson Scientific

• 48 Chemicals, materials and instrumentation
• 14 Telecom, IT, and electronics
• 13 Food and agriculture
• 11 Automotive and transport
• 10 Pharmaceutical and medical
• 4 Energy and power
• Biotechnology Spanish research highly rated in
  agro-industry, medicine, and alternative fuels
• Spanish biotechnology is growing 4 times faster
  than the average of the European 15
• Spain accounts for 4 of all biotech research
  published in the world
• Sluggish integration of IT solutions into daily
  life

7
Integrative Research in Knowledge Management

PEOPLE
PROCESS
Building and Nurturing a Knowledge Sharing Culture
Systematically Capturing and Sharing Critical
Knowledge
TECHNOLOGY
Creating a Unified Knowledge Network
8
Applying AI to KMExpert Systems Technology

• Knowledge elicitation techniques to acquire
  lessons learned (via structured/unstructured
  interviews, protocol analysis, etc.)
• On-line pools of expertise (rule or case-based)
• Knowledge representation techniques for
  developing an ontology

9
Intelligent Agent Technology

• Intelligent multi-agent systems with learning
  capabilities to help users in responding to their
  questions
• Searching and filtering tools
• User profiling and classification tools
• Agent-Oriented Knowledge Management AAAI
  Symposium (Stanford University)

10
Data Mining and Knowledge Discovery Techniques

• Inductively determine relationships/rules for
  further developing the KM system
• Help deduce user profiles for better targeting
  the KM system
• Help generate new cases

11
Neural Networks, Genetic Algorithms, etc.

• Help weed out rules/cases
• Help look for inconsistencies within the
  knowledge repository
• Help filter noisy data

12
KM Research Issues

• --Develop active analysis and dissemination
  techniques for knowledge sharing and searching
  via intelligent agent technology (i.e., where
  learning takes place)
• --Apply knowledge discovery techniques (e.g.,
  data/text mining, neural networks, etc.) for
  mining knowledge bases/repositories
• --Improve query capabilities through natural
  language understanding techniques
• --Develop metrics for measuring value-added
  benefits of knowledge management
• --Develop standardized methodologies for
  knowledge management development and knowledge
  audits
• --Provide improved techniques for performing
  knowledge mapping and building knowledge
  taxonomies/ontologies

13
KM Research Issues (cont.)
--Develop techniques for building collaborative
knowledge bases --Develop improved tools for
capturing knowledge from various media (look at
multimedia mining to induce relationships among
images, videos, graphics, text, etc.) --Develop
techniques for integrating databases to avoid
stovepiping, functional silos --Build improved
software tools for developing and nurturing
communities of practice --Develop techniques for
categorizing, synthesizing, and summarizing
lessons learned (look at text summarization
techniques) --Explore ways to improve human-agent
collaboration --Explore human language
technologies for KM (input analysis, extraction,
question-answer, translation, etc.)
14
WBM 2005 Research Problem (James Simien, NPRST,
April 2005)

• How to provide IT support for the Navys future
  distributed business processes involving sailors
  and commands as outlined in the Navys Human
  Capital Strategy?
• Distributed processes provide tremendous
  opportunity for increasing efficiencies across
  the enterprise.
• Proposed solution
• Develop a Multi-Agent System incorporating
  software agents to intelligently assist Users in
  performing tasks.

15
Major Focus in FY05 (Simien, 2005)

• Development of a formal methodology for knowledge
  acquisition and management for Navys business
  rules used in the assignment process (Liebowitz
  et al., 2005)
• Exploring use of genetic algorithms in Sailor job
  matching
• Development of agent bi-lateral negotiation for
  those assignment matches that occur outside of
  the general matching process
• Experimentation with multiple forms of
  distributed architecture to determine performance
  and scalability (Liebowitz et al., 2004 2005)

16
Next Generation of Data Mining Applications (M.
Kantardzic J. Zurada, IEEE Press, 2005)

• Current data warehouses in the terabyte range
  (FedEx, UPS, Wal-Mart, Royal Dutch/Shell Group,
  etc.)
• Diversity of data (multimedia data)
• Diversity of algorithms (GAs, fuzzy sets, etc.)
• Diversity of infrastructures for data mining
  applications (web-based services and grid
  architectures)
• Diversity of application domains (Internet-based
  web mining, text mining, on-line images and video
  stream mining)
• Emphasis on security and privacy aspects of data
  mining (protect data usually in a distributed
  environment)

17
Red Light Cameras and Motor Vehicle Accidents
(Solomon, Nguyen, Liebowitz, Agresti, 2005
funded through GEICO Found.)

• Objective
• Employ data mining techniques to explore the
  relationship between red light cameras and motor
  vehicle accidents
• Data
• FARS database
• 2000 2003 in MD and Washington, D.C.
• 16,840 entries

18
Findings

• Strongest relationships are collisions with
  moving objects and angle front-to-side crashes.
• The 3pm 4pm hour and months later in the year.
• Car collisions are more likely to happen on
  Fridays and Sundays.
• Types of car crashes involved in running red
  lights are mostly rear-end crashes and angle
  front-to-side collisions.
• High relative importance of gender.

19
New/Repackaged Growth Areas for AI

• Business rule engines
• The acquisition of RulesPower assets allows Fair
  Isaac's customers a higher-performance business
  rule engine (BRE) option that leverages the RETE
  III algorithm (September 27, 2005 Gartner Group
  Report).
• Annual Business Rules Conference (November 2006
  in Washington, D.C.)

20
Another Area for Growth

• Strategic Intelligence The Synergy of Knowledge
  Management, Business Intelligence, and
  Competitive Intelligence (see Liebowitz, J.,
  Strategic Intelligence book, Auerbach
  Publishing/Taylor Francis, NY, April 20, 2006)

21
Continued Growth in Discovery Informatics
(Knowledge Discovery)

• New curricula at the undergraduate level at
  College of Charleston (Discovery Informatics),
  Washington Jefferson (Data Discovery), etc.
• New Graduate Certificate in Competitive
  Intelligence (Johns Hopkins University Jay
  Liebowitz, Program Director)
• SCIP (Society of CI Professionalswww.scip.org)CI
  analysts
• Web and Text Mining

22
Steady Growth

• Robotics and Computer Vision
• Natural Language and Speech Understanding
• Neural Networks, Genetic Algorithms,
  Self-Organizing Maps
• Intelligent/Multi-Agents
• Fuzzy Logic

23
Papers Are Being WrittenWorldwide
EXPERT SYSTEMS WITH APPLICATIONS is a refereed
international journal whose focus is on
exchanging information relating to expert and
intelligent systems applied in industry,
government, and universities worldwide. Published
by Elsevier Entering Volumes 30 31 (2006)
24
Trends in Intelligent Scheduling Systems

• Constraint-based
• Expert scheduling system shells/generic
  constraint-based satisfaction problem solvers
• Object/Agent-oriented, hierarchical architectures
• Hybrid intelligent system approaches

25
NASA Scheduling Environment

• Two of the most pressing tasks in the future for
  NASA Data capture/analysis and scheduling

26
GUESS (Generically Used Expert Scheduling System)

• A generic intelligent scheduling tool to aid the
  human scheduler and to keep him/her in the loop
• Programmed in Visual C and runs on an IBM PC
  Windows environment (about 9,500 lines of code)
• 2.5 year effort

27
Features of GUESS

• OOPS feature of GUESS is that classes represent
  various abstractions of scheduling objects, such
  as events, constraints, resources, etc.
• Resources--binary, depletable, group, etc.
• Constraints--before, after, during, notduring,
  startswith, endswith, meta, etc.
• Repair-based scheduling

28
Major Scheduling Approaches in GUESS

• Suggestion Tabulator uses suggestions derived
  from the constraints
• Hill climbing algorithm
• Genetic algorithm--used EOS, a C class library
  for creating GAs
• Hopfield neural network algorithm

29
Neural Networks in Scheduling

• The existing work demonstrated that scheduling
  problems can be attacked and appropriately solved
  by NNs
• The majority of the artificial NNs proposed for
  scheduling were based on the Hopfield network (an
  optimizer)
• Most of the neural networks developed for
  scheduling have been in manufacturing domains

30
Hopfield Network (NN Connections)

• Each of the constraints on an event produces an
  error signal. The error signal is chosen to cause
  the event to move in the correct direction to
  produce a "satisfied" schedule. The errors on a
  given event induced by the constraints are summed
  together and then passed through a sigmoid
  function. The output of the sigmoid function f(x)
  is used to shift the begin and end times of the
  event to drive the schedule to a more satisfied
  state. Several different sigmoid functions were
  tried. The most promising was f(x) tanh (x).
  This yielded the following equation for the
  neural network

31
Equation Used for NN Connections
32
(No Transcript)
33
Different Types of Scheduling Applications Using
GUESS

• City of Rockville Baseball Scheduling
• Army strategic problem of scheduling arrival of
  units in a deployed theater
• Army operational problem of scheduling Army
  battalion training exercises
• College course timetabling at MC
• NASA satellite scheduling

34
(No Transcript)
35
(No Transcript)
36
Lessons Learned

• Dont underestimate the amount of time required
  for the user interface design
• Scheduling is a difficult (but pervasive) problem
• Nothing goes according to schedule--so have
  efficient ways of handling rescheduling

37
Future Work

• Develop database links for ease of inputting
• Classify different scheduling types and models
  and incorporate them into GUESS
• Expand the number of scheduling methods (ORAI,
  etc.)

38
Questions to Ponder??

• Will AI ever achieve natural/human intelligence?
• Should we have called our field IA (Intelligence
  Amplification) versus AI, since most of the AI
  applications are still for decision support?
• Have we found the killer application for AI
  yet?
• Will AI survive as a field or discipline?

39
THE END

• GRACIAS!!