ICT619 Intelligent Systems

Document Sample
ICT619 Intelligent Systems Powered By Docstoc
					ICT619 Intelligent

Unit Coordinator:
Graham Mann
Room 2.061 ECL Building
Phone: 9360 7270
Unit aims
  to be aware of the rationale of the artificial intelligence
   and soft computing paradigms with their advantages
   over traditional computing

  to gain an understanding of the theoretical foundations
   of various types of intelligent systems technologies to a
   level adequate for achieving objectives as stated below

  to develop the ability to evaluate intelligent systems,
   and in particular, their suitability for specific

  to be able to manage the application of various tools
   available for developing intelligent systems
Unit delivery and learning
  3 hours of lecture/workshop per week

  Lecture/WS time will be spent discussing the relevant
   topic after an introduction by the lecturer

  Topic lecture notes will be available early in the week

  Students should make use of the topic reading material
   in advance for the topic to be covered

  Bringing up issues and questions for discussion are
   encouraged to create an interactive learning
   environment (this is assessed).
Resources and Textbooks
  Main text:
    Negnevitsky, M. Artificial Intelligence: A Guide
     to Intelligent Systems, 2005. 2nd Edition.

  The main text to be supplemented by chapters/articles
   from other books/journals/magazines as well as notes
   provided by the unit coordinator.

  A list of recommended readings and other resources
   will be provided for each topic.

  Unit website: will
   enable access to unit reading materials and links to
   other resources.                                       4

  ACTIVITY        DUE                WEIGHT
  Workshop        Continuous         10%

  Project         Week 12            35%

  Closed-book     Nov exams period   55%

Topic schedule

  Topic 1:     Introduction to Intelligent Systems:
                Tools, Techniques and Applications
    Topic 2:   Rule-Based Expert Systems
    Topic 3:   Fuzzy Systems
    Topic 4:   Neural Computing
    Topic 5:   Genetic Algorithms
    Topic 6:   Case-based Reasoning
    Topic 7:   Data Mining
    Topic 8:   Intelligent Software Agents
    Topic 9:   Language Technology

Topic 1: Introduction to Intelligent
  What is an intelligent system?
  Significance of intelligent systems in business
  Characteristics of intelligent systems
  The field of Artificial Intelligence (AI)
  The Soft Computing paradigm
  An Overview of Intelligent System Methodologies
     Expert Systems
     Fuzzy Systems
     Artificial Neural Networks
     Genetic Algorithms (GA)
     Case-based reasoning (CBR)
     Data Mining
     Intelligent Software Agents
     Language Technology                            7
What is an intelligent system?

  What is intelligence?
     Hard to define unless you list characteristics eg,
         Reasoning
         Learning
         Adaptivity

  A truly intelligent system adapts itself to deal with
   changes in problems (automatic learning)

  Few machines can do that at present

  Machine intelligence has a computer follow problem
   solving processes something like that in humans

  Intelligent systems display machine-level intelligence,
   reasoning, often learning, not necessarily self-adapting
Intelligent systems in business
  Intelligent systems in business utilise one or more intelligence
   tools, usually to aid decision making
  Provides business intelligence to
      Increase productivity
      Gain competitive advantage
  Examples of business intelligence – information on
      Customer behaviour patterns
      Market trend
      Efficiency bottlenecks
  Examples of successful intelligent systems applications in
        Customer service (Customer Relations Modelling)
        Scheduling (eg Mine Operations)
        Data mining
        Financial market prediction
        Quality control
Intelligent systems in business –
some examples
  HNC (now Fair Isaac) software’s credit card fraud detector Falcon
   offers 30-70% improvement over existing methods (an example of
   a neural network).

  MetLife insurance uses automated extraction of information from
   applications in MITA (an example of language technology use)

  Personalized, Internet-based TV listings (an intelligent agent)

  Hyundai’s development apartment construction plans FASTrak-
   Apt (a Case Based Reasoning project)

  US Occupational Safety and Health Administration (OSHA uses
   "expert advisors" to help identify fire and other safety hazards at
   work sites (an expert system).
Characteristics of intelligent
  Possess one or more of these:
       Capability to extract and store knowledge
       Human like reasoning process
       Learning from experience (or training)
       Dealing with imprecise expressions of facts
       Finding solutions through processes similar to natural evolution

  Recent trend
     More sophisticated Interaction with the user through
          natural language understanding
          speech recognition and synthesis
          image analysis

  Most current intelligent systems are based on
     rule based expert systems
     one or more of the methodologies belonging to soft computing

The field of Artificial Intelligence (AI)
  Primary goal:
     Development of software aimed at enabling machines to solve
      problems through human-like reasoning

  Attempts to build systems based on a model of knowledge
   representation and processing in the human mind

  Encompasses study of the brain to understand its structure and

  In existence as a discipline since 1956

  Failed to live up to initial expectations due to
     inadequate understanding of intelligence, brain function
     complexity of problems to be solved

  Expert systems – an AI success story of the 80s

  Case Based Reasoning systems - partial success                   12
The Soft Computing (SC) paradigm
    Also known as Computational Intelligence

    Unlike conventional computing, SC techniques
     1. can be tolerant of imprecise, incomplete or corrupt input data

     2. solve problems without explicit solution steps

     3. learn the solution through repeated observation and

     4. can handle information expressed in vague linguistic terms

     5. arrive at an acceptable solution through evolution
The Soft Computing (SC) paradigm

  The first four characteristics are common in
   problem solving by individual humans
  The fifth characteristic (evolution) is common in
  The predominant SC methodologies found in
   current intelligent systems are:
    Artificial Neural Networks (ANN)
    Fuzzy Systems
    Genetic Algorithms (GA)
Overview of Intelligent System
- Expert Systems (ES)

  Designed to solve problems in a specific domain,
     eg, an ES to assist foreign currency traders

  Built by
     interrogating domain experts
     storing acquired knowledge in a form suitable for solving
      problems, using simple reasoning

  Used by
     Querying the user for problem-specific information
     Using the information to draw inferences from the knowledge
     Supplies answers or suggested ways to collect further inputs

Overview of Expert Systems (cont’d)

  Usual form of the expert system knowledge
   base is a collection of IF … THEN … rules
  Note: not IF statements in procedural code
  Some areas of ES application:
    banking and finance (credit assessment, project
    maintenance (diagnosis of machine faults)
    retail (suggest optimal purchasing pattern)
    emergency services (equipment configuration)
    law (application of law in complex scenarios)
Artificial Neural Networks (ANN)
  Human brain consists of 100 billion densely interconnected simple
   processing elements known as neurons

  ANNs are based on a simplified model of the neurons and their

  ANNs usually learn from experience – repeated presentation of
   example problems with their corresponding solutions

  After learning the ANN is able to solve problems, even with newish

  The learning phase may or may not involve human intervention
   (supervised vs unsupervised learning)

  The problem solving 'model' developed remains implicit and
   unknown to the user

  Particularly suitable for problems not prone to algorithmic
   solutions, eg, pattern recognition, decision support
Artificial Neural Networks (cont’d)

  Different models of ANNs depending on
      Architecture
      learning method
      other operational characteristics (eg type of activation function)

  Good at pattern recognition and classification problems

  Major strength - ability to handle previously unseen, incomplete or
   corrupted data

  Some application examples:
     - explosive detection at airports
     - face recognition
     - financial risk assessment
     - optimisation and scheduling

Genetic Algorithms (GA)

  Belongs to a broader field known as evolutionary computation

  Solution obtained by evolving solutions through a process
   consisting of
      survival of the fittest
      crossbreeding, and
      mutation

  A population of candidate solutions is initialised (the

  New generations of solutions are produced beginning with the
   intial population, using specific genetic operations: selection,
   crossover and mutation

Genetic Algorithms (cont’d)
 Next generation of solutions produced from the current population
     crossover (splicing and joining peices of the solution from parents) and
     mutation (random change in the parameters defining the solution)

 The fitness of newly evolved solution evaluated using a fitness

 The steps of solution generation and evaluation continue until an
  acceptable solution is found

 GAs have been used in
       portfolio optimisation
       bankruptcy prediction
       financial forecasting
       design of jet engines
       scheduling
Fuzzy Systems
  Traditional logic is two-valued – any proposition is
   either true or false

  Problem solving in real-life must deal with partially true
   or partially false propositions

  Imposing precision may be difficult and lead to less
   than optimal solutions

  Fuzzy systems handle imprecise information by
   assigning degrees of truth - using fuzzy logic

Fuzzy Systems (cont’d)

 FL allow us to express knowledge in vague linguistic

 Flexibility and power of fuzzy systems now well
  recognised (eg simplification of rules in control systems
  where imprecision is found)

 Some applications of fuzzy systems:
    Control of manufacturing processes
    appliances such as air conditioners, washing machines and
     video cameras
    Used in combination with other intelligent system
     methodologies to develop hybrid fuzzy-expert, neuro-fuzzy,
     or fuzzy-GA systems

Case-based reasoning (CBR)

  CBR systems solve problems by making use of knowledge about
   similar problems encountered in the past

  The knowledge used in the past is built up as a case-base

  CBR systems search the case-base for cases with attributes
   similar to given problem

  A solution created by synthesizing similar cases, and adjusting to
   cater for differences between given problem and similar cases

  Difficult to do well in practice, but very powerful if you can do it

Case-based reasoning (cont’d)

 CBR systems can improve over time by learning from
  mistakes made with past problems

 Application examples:
      Utilisation of shop floor expertise in aircraft repairs
      Legal reasoning
      Dispute mediation
      Data mining
      Fault diagnosis
      Scheduling

Data mining

  The process of exploring and analysing data for
   discovering new and useful information

  Huge volumes of mostly point-of-sale (POS) data are
   generated or captured electronically every day, eg,
     data generated by bar code scanners
     customer call detail databases
     web log files in e-commerce etc.

  Organizations are ending up with huge amounts of
   mostly day-to-day transaction data

Data mining (cont’d)
  It is possible to extract useful information on market and customer
   behaviour by “mining" the data

  Note: This goes far beyond simple statistical analysis of numerical
   data, to classification and analysis of non-numerical data

  Such information might
      reveal important underlying trends and associations in market
       behaviour, and
      help gain competitive advantage by improving marketing

  Techniques such as artificial neural networks and decision trees
   have made it possible to perform data mining involving large
   volumes of data (from "data warehouses").

  Growing interest in applying data mining in areas such direct
   target marketing campaigns, fraud detection, and development of
   models to aid in financial predictions, antiterrorism systems
Intelligent software agents (ISA)
 ISAs are computer programs that provide active assistance to
  information system users

 Help users cope with information overload

 Act in many ways like a personal assistant to the user by
  attempting to adapt to the specific needs of the user

 Capable of learning from the user as well as other intelligent
  software agents

 Application examples:
     News and Email Collection,
      Filtering and Management
     Online Shopping
     Event Notification
     Personal scheduling
     Online help desks, interactive characters
     Rapid Response Implementation                                27
Language Technology (LT)
 “[The] application of knowledge about human language in computer-
  based solutions” (Dale 2004)

 Communication between people and computers is an important
  aspect of any intelligent information system

 Applications of LT:
      Natural Language Processing (NLP)
      Knowledge Representation
      Speech recognition
      Optical character recognition (OCR)
      Handwriting recognition
      Machine translation
      Text summarisation
      Speech synthesis                        Hi, I am Cybelle.
                                              What is your name?

 A LT-based system can be the front-end of
  information systems themselves based on
  other intelligence tools                                         28
For Next Week

 Get hold of the textbook
 Visit the library and find the section on
  artificial intelligence, browse some titles
 Get onto the unit website, download and
  read papers concerning Expert Systems
 We will study the theory and practice
  developing a simple expert system
 Have a look at the AAAI Applications
  webpage at