NEURAL NETWORKS

1
NEURAL NETWORKS

• Fuzzy logic

2

• Neural network is dubbed as an replica of a
  human brain on the lines of its working model
• The functionality of neural network can be
  equated to the functions of an human brain
• How does human brain learn
• 1. the human brain transforms the given input to
  output is the general idea
• (inputs)
  (outputs)

3

• The human brain consists of small elements
  called Neurons
• 1.The neurons collects the signal through a
  host of fine
  structures called dendrites
• 2. the neurons then sends out the electrical
  activity
• through a thin stand called Axons

4

• 3. These axons are then split into thousands of
  branches
• 4.At the end of each branch there is a structure
  called synapse
• 5.This synapse sends the electrical activity to
• other neurons which are interconnected

5

• When the recipient neuron traces out that the
  input is large in size in comparison to its
  original size, its sends the electrical activity
  down the axon
• Synapses play an significant role in transferring
  the data from one neuron to another
• So, by changing the effectiveness of the synapses
  learning occurs
• A neural network also functions in the exact
  style
• A neural network is fused up by a series of small
  elements called neurons

6

• We can train a neural network to perform a
  particular task
• The fancy point about a neural network is it can
  be adjusted and trained so that the input
  leads to the specific target of output
• Hence neural n/w is also called as artificial
  neural n/w
• This is called supervised learning like a
  learning of a human brain
• The human brain generated output based on the
  inputs given

7

• But the neural network is a good adjustor of
  neurons and the desired and target result can be
  outputted
• BASIC STRUCTURE OF A NEURAL NETWORK
• 
  
  
  
  hidden layer
• Input layer
  
  output layer
• 
  
• 
  
  
• 
  
  
  
• 
  
• 
  
• 
  
• 
  
• 
  (weights ) synapses
  
  
  
  

8

• BLOCK DIAGRAM OF NEURAL N/W
• 
  
  
  TARGET
• INPUTS O/P
  
• 
  
  OUTPUTS
• 
  ADJUST WEIGHTS
• The o/p is not matched with the target o/p the
  weights can be adjusted, this particular
  flash-point as made neural network a remarkable
  tool

NEURAL NETWORK INCLUDING
CONNECTIONS CALLED WEIGHTS
COMPARE
9

• The connection between neurons are called weights
  the
• Weight values are adjusted to get the target
  output.
• Take a single neuron, in a n/w it has two modes
  of operation
• 1.Training mode
• 2.Firing mode
• in training mode the neurons will be trained to
  fire for a particular input patterns
• In the firing mode the neuron has two tasks
• To fire if the given input is form the trained
  list of input patterns/ fire in case of any
  similarities
• 2.vice-versa
• Lets take a best example of a 3-input neuron
• X1,x2,x3 are three neurons

10

• The neuron here is trained in such a style so as
  to
• Case1output 0(dont fire) if the input is
  111(or)101
• Case2Output1 (fire) if the input is 000 (or) 001

X1 0 0 0 0 1 1 1 1
X2 0 0 1 1 0 0 1 1
X3 0 1 0 1 0 1 0 1
O/P 0 Case1 (not fire) 0 Case1 0/1 None of the above 0/1 None of the above 0/1 None of the above 1 Case2 (neuron to fire) 0/1 None of the above 1 Case2
(after firing rule) 0 0 0 0/1 0/1 1 1 1
11

• Consider the third column which is 010, which is
  before undefined after firing outputs the value
  0 HOW?

12

• Lets consider fourth column which is before 0/1
  after applying the firing rule also holds the
  constant value,how!!!! HOW?
• undefined, after applying firing rule it is
  0,HOW?

CONSIDER 011 COMPARE AND CONTRAST WITH ALL THE FOUR SETS Case1 111 output is 1 Case 1 101 Output is 1 Case 2 000 Output is 0 Case 2 001 Output is 0
011 X21 X31 DIFFERS ONLY ONE ELEMENT X31 DIFFERS IN TWO ELEMENTS X10 DIFFERS IN TWO ELEMENTS X10 X31 DIFFERS IN ONLY ONE ELEMENT
OUTPUT AFTER FIRING (MAXIMUM SIMILARITY) HENCE TAKE THE VALUE OF 111CASE1) 1 (MAXIMUM SIMILARITY) HENCE TAKE THE VALUE OF 001CASE2) 0
13

• Because it hold maximum similarities with both
  the cases( case1,case2)
• The firing rule states that it has to remain
  undefined because of a tie
• with the same mechanism of neurons getting
  trained/adjusted/fired, and outputting the target
  o/p has made neural n/w instrumental in a many
  spheres
• Neural network merged with fuzzy logic ha done
  wonders in the fields of data mining ETC

14

• Fuzzy logic tool was introduced in 1965 by lot
  fi zadeh
• Fuzzy means something which is blurred/ hazy
• Fuzzy logic means is a mathematical tool that
  deals with uncertainty
• Haziness persist in any realistic process, fuzzy
  logic task is to decode exactness out of
  something which is inexact
• The human brain has the capability to make a
  clear distinction between an image and an object
  even if it is blur
• Linear computing is able to read just pixels as a
  set of colours
• Fuzzy logic capability to solve problems that
  linear computing is not able to do.

15

• Fuzzy logic hence embedded in neural networks
  show more transparency
• APPLICATIONS of neural networks
• speech recognition
• Pattern recognition
• image processing
• data mining
• robotics
• data segmentation and compression

16

• Fuzzy logic is used to model systems that has
  ambiguity or opaqueness ,it can be vagueness/lack
  of information/miscalculation of measurements
• EXAMPLE
• Entity x to this entity a short person may be
  one whose height is below 4.20
• Entity y to this entity a short person may be
  one whose height is beneath or equal to 3.9
• Here short is the language descriptor , it
  applies the same meaning to both x and y but it
  established that they dont have a unique
  definition for short
• Such type of information associated with dilemma
  are made feasible to the computers with the tool
  called fuzzy logic

17

• The fuzzy logic incorporates a simple IF x AND
  y THEN z approach rather than modeling a system
  mathematically
• Example
• Rather than Dealing the temperature control in
  terms such as 1. SP500f
• 2.Tlt1000f
• 3.210cltTEMPlt220c
• Fuzzy logic deals in terms like
• IF(process is too cool) AND( getting colder)
  THEN(add heat to the process)

18

• 2. IF(process is too hot) AND (process is heating
  rapidly) THEN (cool the process quickly)
• Because of this potential to deal with complex
  tasks fuzzy logic has wide range of application
  having its share in all household appliances
• 1.Washing machines
• 2. Electric rice cookers
• 3.Speech recognition
• 4.Stock market predictions
• 5.High speed trains

19

• fuzzy logic in washing machines
• The washing machine first tests how dirty the
  laundry is
• Once it knows how dirty the laundry is it can
  easily calculate how long it can wash it
• First it always take a base of 10minutes
• Then if the cloth put in it is 100 dirty then it
  adds two minutes (10212 minutes)
• If the cloth put in the washing machine is 50
  dirty then it adds 1 minute to the base of
  10min(10111min)
• The laundry can also be greasy at the same time

20

• If the laundry is greasy then add 2minutes to the
  base of 10min(12min)
• If the laundry is 50 greasy then add 1 minute to
  the base of 10min(11min)
  
• 
  1.
• 
  
• 
  2.
• 
  3.

FUZZ MACHINE
21

• Shirt 1 100 dirty( 2min)
• Shirt 2 100 clean(0 min)
• Shirt 350 greasy(1min)
• Total time taken by the washing machine working
  with fuzz logic is 10(base)2min0min1min
  13min ( for three shirts)

22

• ADVANTAGES OF NEURAL N/W AND FUZZY LOGIC
• High accuracy neural n/w are able to give the
  exact result of complex systems
• Noise tolerance neural n/w are very flexible
  with respect to incomplete, missing and noisy
  data
• Ease of maintenance neural n/w can be updated
  with fresh data making them useful for dynamic
  environments
• When an element (i.e) neuron fails the other
  neuron undertakes the task

23

• Though the advent and discovery of neural network
  is dated back to 1943 by warren mc culloch, it
  has been a wonder tool in networks till date
• Neural networks has been enhanced and able to
  hammer out solutions for the problems which are
  complex for conventional computers/human beings
• Neural network is merited in many ways with only
  one setback i,e training the neurons to generate
  a target o/p
• Hence neural networks and fuzzy logic are
  supplementary to computers