Iris setosa, versicolor and virginica

trainerhungarianAI and Robotics

Oct 20, 2013 (4 years and 21 days ago)

110 views

Problem için uygun akıllı sistem seçimi


Bu sunum
M.
Negnevitsky



Artificial
Intelligence: A Guide to Intelligent Systems,
Addison
-
Wesley, 2002, 2nd
Edition

slaytlarından alınmıştır.

Problem Type

Diagnostic Expert System


I want to develop an intelligent system that
can help me to fix malfunctions of my Mac
computer. Will an expert system work for this
problem?

Diagnostic Expert System

Classification of Expert System


I want to develop an intelligent system that
can help me to identify different classes of
sail boats. Will an expert system work for this
problem?

Classification of Expert System

Certainty Factor

Certainty Factor

Will a Fuzy Expert System Work for My
Problem



if you cannot define a set of exact rules for each

possible situation, then use fuzzy logic


Fuzzy systems are particularly well suited for
modelling human decision making.


most fuzzy technology applications are still reported
in control and engineering, an even larger potential
exists in business and finance


Fuzzy technology provides us with a means of
coping with the ‘soft criteria’ and ‘fuzzy data’ that are
often used in business and finance

Decision
-
support fuzzy systems


I want to develop an intelligent system for
assessing mortgage applications. Will a fuzzy
expert system work for this problem?

Define Fuzy Set

Define Fuzy Set

Define Fuzy Set

Define Fuzy Set

Define Fuzy Set


Hierarchical fuzzy model for mortgage
loan assessment

Three
-
dimensional plots for Rule Base 1
and Rule Base 2

Will a neural network work for my
problem


Neural Network is good for prediction, classification and clustering
problems


Areas


speech and character recognition


detecting fraudulent transactions


medical diagnosis of heart attacks to process control


Robotics


predicting foreign exchange rates to detecting and identifying radar targets


the areas of neural network applications continue to expand rapidly.


The popularity of neural networks is based on their remarkable
versatility, abilities to handle both binary and continuous data, and to
produce good results in complex domains.


When the output is continuous, the network can address prediction
problems, but when the output is binary, the network works as a
classifier

Character recognition neural networks


I want to develop a character recognition
system. Will a neural network work for this
problem?

Bitmap for Digit Recognition

How do we choose the architecture of a
neural network for character

recognition?



The architecture and size of a neural network
depend on the complexity of the problem


handwritten character recognition is performed by
rather complex multilayer networks that may
include three, or even four, hidden layers and
hundreds of neurons


for the printed digit recognition problem, a three
-
layer network with a single hidden layer will give
sufficient accuracy.

Neural network for printed digit
recognition

Input and Desired Output

Traning and Performance Evaluation

Prediction neural networks


I want to develop an intelligent system for real
-
estate appraisal. Will a neural network work for this
problem?


In this problem, the inputs (the house location, living area,
number of bedrooms, number of bathrooms, land size, type
of heating system, etc.)


there are many examples we can use for training the neural
network. These examples are the features of recently sold
houses and their sales prices.


Choosing training examples is critical for an accurate
prediction. A training set must cover the full range of values
for all inputs.

How do we massage the data


For instance, if the living areas of the houses in training examples
range between 59 and 231 square metres, we might set the
minimum value to 50 and the maximum to 250 square metres. Any
value lower than the minimum is mapped to the minimum, and any
value higher than the maximum to the maximum. Thus, a living area
of, say, 121 square metres would be massaged as

Feedforward neural network for real
-
estate
appraisal

But how do we interpret the network
output?


In our example, the network output is represented by continuous values in
the range between 0 and 1. Thus, to interpret the results, we can simply
reverse the procedure we used for massaging continuous data. Suppose,
for instance, that in the training set, sales prices range between $52,500
and $225,000, and the output value is set up so that $50,000 maps to 0 and
$250,000 maps to 1. Then, if the network output is 0.3546, we can compute
that this value corresponds to:

Classification neural networks with
competitive learning


I want to develop an intelligent system that can divide a group of iris
plants into classes and then assign any iris plant to one of these
classes. I have a data set with several variables but I have no idea
how to separate it into different classes because I cannot find any
unique or distinctive features in the data. Will a neural network work
for this problem?


Clusturing (clustering can be defined as the process of dividing an input
space into regions
)


we will use a data set of 150 elements that contains three classes of iris
plants:
Iris setosa, versicolor and virginica


Each plant in the data set is represented by four variables:
sepal length,
sepal width, petal length and petal width



Neural network for iris plant classification

Competitive learning in the neural network for iris plant
classification: (a) initial weights; (b) weight after 100 iterations; (c)
weight after 2000 iterations

Will genetic algorithms work for my
problem


Genetic algorithms are applicable to many
optimisation problems


Optimisation is essentially the process of finding a better
solution to a problem


This implies that the problem has more than one solution
and the solutions are not of equal quality


A genetic algorithm generates a population of
competing candidate solutions and the causes them
to evolve through the process of natural selection


poor solutions tend to die out, while better solutions survive
and reproduce. By repeating this process over and over
again, the genetic algorithm breeds an optimal solution


The travelling salesman problem


I want to develop an intelligent system that can
produce an optimal itinerary. I am going to travel by
car and I want to visit all major cities in Western and
Central Europe and then return home. Will a genetic
algorithm work for this problem?


Researchers apply different techniques to solve this
problem. These techniques include simulated annealing
(Laarhoven and Aarts, 1987), discrete linear programming
(Lawler et al., 1985), neural networks (Hopfield and Tank,
1985), branch
-
and
-
bound algorithms (Tschoke et al.,
1995), Markov chains (Martin et al., 1991) and genetic
algorithms (Potvin, 1996).


Genetic algorithms are particularly suitable for the TSP
because they can rapidly direct the search to promising
areas of the search space.

How does a genetic algorithm solve the
TSP?


Suppose we have nine cities numbered from
1 to 9. In a chromosome, the order of the
integers represents the order in which the
cities will be visited by the salesman. For
example, a chromosome



How does the crossover operator work in
the TSP?

Crossover operators for the TSP

The travelling salesman problem


How does the mutation operator works in the
TSP?


The reciprocal exchange operator simply swaps
two randomly selected cities in the chromosome.
The inversion operator selects two random points
along the chromosome string and reverses the
order of the cities between these points.


Will a hybrid intelligent system work for
my problem?


Solving complex real
-
world problems requires an
application of complex intelligent systems that
combine the advantages of expert systems, fuzzy
logic, neural networks and evolutionary computation


Although the field of hybrid intelligent systems is still
evolving, and most hybrid tools are not yet
particularly effective, neuro
-
fuzzy systems have
already matured as an advanced technology with
numerous successful applications. While neural
networks can learn from data, the key benefit of
fuzzy logic lies in its ability to model decision
-
making
of humans.

Neuro
-
fuzzy decision
-
support systems


I want to develop an intelligent system for
diagnosing myocardial perfusion from cardiac
images. I have a set of cardiac images as
well as the clinical notes and physician’s
interpretation. Will a hybrid system work for
this problem?

Fuzy Data Set

Fuzzy rules for assessing the risk of a heart
decease

Hierarchical structure of the neuro
-
fuzzy system for risk
assessment of the cardiac decease

Implementetion of the diagnostic system