You are on page 1of 18

Advanced Topics in

Artificial Intelligence

1
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Some topics in AI course
z Fuzzy logics
z Neural Networks
z Genetic Algorithms

2
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Fuzzy logics
z Ideas
– Human beings often need to deal with input
that is not in precise or numerical form.

3
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Fuzzy logics
Fuzzy sets
– Zadeh (1965) developed a fuzzy set theory
that allows concepts that do not have well-
defined sharp boundaries.
– Classic sets: an object full member or full
non-member
– Fuzzy sets: partial membership
z membership function

4
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Fuzzy logics
z “if x is A” is partially satisfied if the object
x (usually crisp value x) is partial
membership of the fuzzy set
z “if-then” rules
z When the “if” condition is partially
satisfied, the conclusion of a fuzzy rule is
drawn based on the degree to which the
condition is satisfied

5
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Fuzzy logics
Membership functions

6
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Fuzzy rule-based system (FRBS)
if (x1 is I1,k) and (x2 is I2,k) and ... and (xn is In,k)
then y is Ok (k = 1..m)

m: number of fuzzy rules


x1, x2, ..., xn: inputs
I1,k , I2,k, ..., In,k corresponding fuzzy sets
(membership functions µIi,k)
y: output (Ok - corresponding fuzzy set,
membership function µOk)
7
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Fuzzy rule-based system (FRBS)
Example:

If CurrentTemp is High and FanSpeed is


VeryLow
then
IncreaseFanSpeed is VeryHigh

8
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Fuzzy rule-based system (FRBS)
z Because a crisp value can be partial membership
of several fuzzy sets, it activates a number of
fuzzy rules to some degree
z The extent to which the k-th rule is activated is
calculated as:
αk = µ I1,k (x1) ^ µ I2,k (x2) ^ ::: ^ µ In,k (xn)
z The inferencing membership value of the output
y is calculated as:
µinf Ok = αk ^ µ Ok (y)
min is usually used for ^
9
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Fuzzy rule-based system (FRBS)
z The total membership value of the output y
is calculated by the compositional rule of
inference as:

z max is usually used for ˅

10
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Fuzzy rule-based system (FRBS)
z Defuzzication methods: Center of Area
(COA), Center of Maximum (COM), and
Mean of Maximum (MOM)
z COA:

11
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Genetic Algorithms
z Ideas:
– Natural Evolutional process

12
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Genetic Algorithms
z Search algorithms based on the process of
natural evolution and survival of the fittest
in the biological world (Goldberg, 1989).
z While traditional optimization techniques
search for an optimal solution from a
single point, GAs search from a population
of solutions

13
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Genetic Algorithms
Initialize the population
Evaluate initial population
Repeat
Perform competitive selection
Apply genetic operators to generate new
solutions
Evaluate solutions in the population
Until some convergence criteria is satisfied

14
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Neural Networks
z Ideas:
– Human brain organization
– Human neural networks

15
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Neural Networks
z also known as Artificial Neural Networks
(ANNs), Connectionist Models, and Parallel
Distributed Processing (PDP) Models
z "Artificial Neural Networks' are massively
parallel interconnected networks of simple
(usually adaptive) elements and their hierarchical
organizations which are intended to interact with
the objects of the real world in the same way as
biological nervous systems do." -- T. Kohonen

16
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Neural Networks
z A number of very simple, neuron-like processing
elements called units, PEs, or nodes
z A number of weighted, directed connections
between pairs of units
z Weights may be positive or negative real values
z Local processing in that each unit computes a
function based on the outputs of a limited
number of other units in the network
z Each unit computes a simple function of its input
values, which are the weighted outputs from
other units.

17
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006
Neural Networks
z Perceptron: Simplest "interesting" class of
neural networks
z Feedforward Neural Nets:
– Backpropagation Learning
– Gradient of E

18
Bùi Thế Duy - Bộ môn Mạng và TTMT 6/22/2006

You might also like