Professional Documents
Culture Documents
An Introduction To Hybrid Intelligent Systems: November 2013
An Introduction To Hybrid Intelligent Systems: November 2013
net/publication/312086965
CITATIONS READS
2 2,095
1 author:
Abdullatif Baba
Türk Hava Kurumu Üniversitesi
102 PUBLICATIONS 65 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Genetic algorithm-based technique for predicting future generations of hazelnuts chromosomes. View project
All content following this page was uploaded by Abdullatif Baba on 05 January 2017.
12/22/2016 ababa@thk.edu.tr 1
Hybrid Intelligent Systems
Probabilistic reasoning, fuzzy logic, neural networks and evolutionary
computation.
By combining the advantages of each technology we can create a more powerful and
effective expert system
Hybrid Intelligent Systems
Neural expert systems
• The knowledge base is represented
by a trained neural network.
• The rule extraction unit examines the
neural knowledge base and produces
the rules implicitly in the trained
neural network.
• The explanation facilities explain to
the user how the neural expert system
arrives at a particular solution when
working with the new input data.
• The user interface provides the means
of communication between the user
and the neural expert system
Remember:
The neurons in the second layer apply the following sign activation function
Hybrid Intelligent Systems
Neural expert systems
How does a neural expert system extract rules that justify its inference?
For example, if the object has Wings +1, Beak +1 and Feathers +1, but does not have
Engine -1, then we can conclude :
Hybrid Intelligent Systems
Neural expert systems
How does a neural expert system extract rules that justify its inference?
We can similarly conclude that this object is not Plane,
If a given object has wings and beak and feathers Then it is a bird
If a given object has wings and beak and engine Then it is a plane
.
.
Hybrid Intelligent Systems
Neural expert systems
Can we define a system that obtains the most important information first to draw a conclusion as
quickly as possible?
And thus, according to the heuristic, the following inference can be made:
CONCLUDE: BIRD IS TRUE
Note: UNKNOWN indicates how this net input might change based upon the worst possible
combination of values of the unknown inputs. In our example, the net weighted input cannot change
more than ±2.1. Therefore, the output of neuron Rule 1 will be greater than 0 regardless of the
values of the known inputs.
Hybrid Intelligent Systems
Neural expert systems
how a single rule can be extracted to justify an inference?
Let us justify the inference that Bird is true
But at first, we have to determine all contributing inputs and the size of each
contribution
• All neurons in the first layer are directly connected to neuron Rule 1, we might expect
that the rule to be extracted may involve all five neurons : Wings, Tail, Beak, Feathers
and Engine.
• The size of this contribution is determined by the absolute value of the weight 𝑤𝑖 of
the contributing input i.
In our example, the list of inputs contributing to the inference “Bird is true” looks as follows:
The rule is not valid yet, and thus we need to add the ‘second best’ contributing
input as a clause in the condition part of our rule:
Combined
Neuro-fuzzy systems
A standard neuro-fuzzy system has input and output layers, and three hidden layers that
represent membership functions and fuzzy rules
Two inputs x1 and x2. One output y. Input x1 is represented by three fuzzy sets A1, A2 and
A3; input x2 by three fuzzy sets B1, B2 and B3; and output y by two fuzzy sets C1 and C2.
Neuro-fuzzy systems
Layer 2 is the input membership or fuzzification layer:
And so on …
Neuro-fuzzy systems
• The weights between Layer 3 and Layer 4 have to be normalised.
They are selected randomly and during training, they can be
changed. (Confidence degree attributed to each rule)
To keep them with in the specified range , the weights are normalised
by dividing their respective values by the highest weight magnitude
obtained at each iteration.
Combined
Hybrid Intelligent Systems
Evolutionary neural networks
Example:
• Initial weights in the network are chosen randomly within some small interval [-1,1]
• Each row in the last matrix represents a group of all the incoming weighted links to a
single neuron. This group can be considered as a functional building block of the
network. (Thus, each gene is a group of all incoming weights of a given neuron)
Hybrid Intelligent Systems
Evolutionary neural networks
Example (Crossover) :
-
Hybrid Intelligent Systems
Evolutionary neural networks
Example (Mutation) :