Introduction to Artificial Intelligence, Soft Computing, their

Document Sample
Introduction to Artificial Intelligence, Soft Computing, their Powered By Docstoc
					Introduction to Artificial Intelligence,
 Soft Computing, A Case Study and
         Future Implication

                             by

                       K. Lavangnananda
            School of Information Technology (SIT)
   King Mongkut’s University of Technology Thonburi (KMUTT)


                                          Sunday 19th October 2008.
                                         Graduate School of Computer
                                          Assumption University
              Definitions of
       Artificial Intelligence (AI)

The study of mechanisms that think and act like
                        humans
       -------------------------------------------
The study of mechanisms underlying intelligent
     behaviour through the construction and
     evaluation of artifacts that enact those
                     mechanisms
         Is
‘machine intelligence’
     possible ?
Concepts/Definition
  of ‘Intelligence’
Can machine
  think ?
    Introduction to Soft Computing /
       Computational Intelligence
Many believe that this is a modern approach to AI.
At present, there is no precise definition of these
  terms.
However, techniques in Soft Computing /
  Computational Intelligence are :
  Fuzzy Logic
  Evolutionary Computation
  Neural Networks
  (Probabilistic Reasoning ?)
An example of Evolutionary Computation in
Knowledge Discovery (a data mining program
                 SARG)

 SARG is an acronym for Self-adjusting Association Rules
 Generator
 An evolutionary computation system known, based on genetic
 programming known as Self-adjusting Association Rules
 Generator (SARG) was implemented.
 SARG comprises 3 main components
  – Data preprocessing
  – Evolutionary computation
  – Final rule builder
 Data preprocessing
• The data set must be split into 2 sets, training set and test set.

 Evolutionary Computation
Evolutionary computation
Final Rule Builder

The format of the final classification rule is :
IF [condition(s) for the rule with highest fitness value]
THEN (class = category of the rule with highest fitness value)
ELSE IF [condition(s) for the rule with 2nd highest fitness
           value]
       THEN (class = category of the rule with 2nd highest
               fitness value)
               ……….
       ELSE IF [condition(s) for the rule with lowest fitness
                   value]
               THEN (class = category of the rule with lowest
                        fitness value)
                ELSE (sample is unclassified)
An Example : Predicting M.Sc. IT
       students’ GPA
The range of GPA is between 0 and 4.
After detailed analysis of student files, eight
measurable attributes were considered relevant
in judging whether an applicant should be
admitted to the programme. These are shown in
the following table.
Degrees of success (i.e. GPA expected) were
classified into 3 categories
      >= 3.5 (Class 1)
      3.0 - 3.5 (Class 2)
      <= 3.0 (Class 3)
              Attribute                                  Value
A1 (Gender)                            1 (Male) ; 2 (Female)
A2 (Age)                               1 (20-25) ; 2 (26-30) ;
                                       3 (31-35) ; 4 (36-40) ; 5 (over 40)

A3 (Marital Status)                    1 (Single) ; 2 (Married)
A4 (Qualification)                     1 (1st Degree) ; 2 (2nd Degree)
A5 (Subject Area in 1st degree)        5 different categories
                                       were considered relevant

A6 (Type of Institute in 1st degree)   5 different categories
                                       were considered relevant

A7 (GPA in 1st degree)                 1 (3.5 – 4) ; 2 (3.0 – 3.5) ;
                                       3 (2.5 – 3.0) ; 4 (under 2.5)

A8 (Employment Status)                 1 (Employed) ; 2 (unemployed)
              Datasets used

 Data set available consisted of 276 past
student records. They were taken from past
student files from semester 2/1996 to semester
1/1999.
  Training set consisted of 200 samples while
76 samples were set aside for testing.
  After numerous experiments, SARG yielded
the best performance of 81.16% accuracy and
produced 6 rules.
                Points to note

The limited number of samples available for training and
testing may be crucial : the task may be too difficult if
sufficient number of samples is not available for training
The maximum number of conditions allowed in a rule has a
direct influence on performance : The maximum number in
this work was set to 3 to avoid rules becoming to specific to
the training set. This may be insufficient too. However, setting
this number higher will make the task of generating rules
much harder since more and longer chromosomes will be
required as well as more iterations for each number of
conditions.
Quality of attributes is another crucial factor : Selecting
relevant attributes requires careful analysis indeed. In this
work, quality attributes such as ‘no. of hours spent on revision
each week’ and ‘relevant experience’ cannot be obtained
easily or almost impossible to assess.
     Future implication of AI
    Computer and IT technology have come
to the point that the improvement and the
future potential of machines and devices do
not lie in their ability to do mundane tasks or
processing data. People are, more and more,
expecting these machines and devices to
perform some decision making.
The above can be translated to the
   need to program computing
devices not what to do but how to
  do. AI is such a discipline that
  provide the basis for this need.
Hence, the success in fulfilling the
 requirement of future computing
 devices lies in the success in AI
             research.
      Future implication of AI
    The benefit from ‘intelligent’ machines are
plentiful. They can assist, or even replace human in
performing tasks which require intelligence. The
ultimate goal of AI was clear from the beginning. It
was meant to improve the quality of human life and to
the betterment of society as a whole.
    However, the implication of AI is not quite as clear
as its goal. This has been an on-going debate for
sometimes. The main issue is to what extent should
human allow the decision making process to machine
?
                           .
So far, the impact of AI has far
more positives than negatives.
Thank you for your attention
  Questions are welcome