Ali Setoodehnia Ph.D.

1

SHORT TALK ON ARTIFICIAL NEURAL NETWORK
(ANN) Research Classroom Lecture on Error
Detection

• Ali Setoodehnia Ph.D.
• Kean University
• Union, NJ 07083
• asetoode_at_kean.edu
• 908-737-3507

2

• AGENDA
• PART-IANN
• Introduction
• Neural Network
• Learning Process
• Feed-forward Network
• Local Recurrent Network
• Convergence
• Simulations
• Applications
• Summary
• PART-II Classroom Lecture

3
Introduction

• Who am I?
• Why are we here?
• Ask me question?

4
Artificial Neural Network (ANN)

• What is ANN?
• ANN is parallel processing technique which is a
  new form of solution for nonlinear systems.
• ANN is MLIN with SLR for updating the connection
  weights.
• Why ANN?
• Conventional technique are very successful in
  some area like linear system, but for nonlinear
  systems there is question.

5
Ann Learning Process

• A typical ANN learning process is based on the
  following characteristics
• Architecture
• Learning mechanism
• Learning modes
• Learning rate
• Learning law

6
Architecture

• FFMN
• Modified-FFMN
• GMDH

7
Feed-forward MULTILAYER Network (FFMN)

• What is FFMN?
• A FFMN is a net with one or more layers of
  neurons between inputs and the output units ,
  which the signals flow from the input units to
  the output units, in a forward direction.

8
FFMN Architecture
9
XOR - Models

• Single neuron

x2
x1
(1,1)
(0,1)
output
(1,0)
(0,0)
x1
x2
10
XOR-models

• Two layer

x2
x2
(1,1)
(0,1)
OUTPUT
x1
(1,0)
(0,0)
x1
11
XOR-models

• Three layer

x2
x2
x1
OUTPUT
(1,1)
(0,1)
(1,0)
(0,0)
x2
x1
x1
12
Learning Rule

• Hebb
• Perceptron
• ADALIN
• BackPropagation

13
HEBB Rule

• If both x(input) and y(output) are active (ON)
  then w(t1) w(t) XY
• This is good for logic functions and character
  recognition
• Data representation can be
• Binary 0,1
• Bipolar 1, 0, -1

14
PERCEPTRON Rule

• If error (input - output) is not zero
  
• Then
• W(t1) w(t) µ XY
• Where µ learning rate

15
ADELIN Rule

• If the error in not less than threshold
• Then
• W(t1) w(t) µ (target - output_ measur.)

16
BACKPROPAGATION
error
dk

• Path.
• If error is not less than threshold then for the
  hidden layers the weights are updated as
• ?Wij(t) µ(Zj)(Sk((Zk)ek)Yi
  ß?Wij(t-1).
• Where ß is momentum factor, and f is tanh()
  function or sigmoid function.
• And for output layer
  ?Wjk(t) µ(Zk)(dk - Yk)Yj
  ß?Wjk(t-1).

j
k
i
Wij
Wik
Yk
17
(No Transcript)
18
Modified FFMN

• Installing ARMA filter at each neuron
• Convergence

19
(No Transcript)
20
(No Transcript)
21
Single Neuron Proof

22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
Client-Server

• CLIENT APPLET

Text/GUI
Raw Data
Internet
MAIN SERVER
Intranet
LOCAL SERVER
27
VCL Simulation

• The following procedure is used for VCL and
  running experiment
• Step-1 Get access to the Internet browser
• Step-2 Type VCLs URL address
• Step-3 Choose an option
• Step-4 Enter the data in the Text Fields
• Step-5 Click Run
• Step-6 see the plot for Convergence check!
• Step-7 Save the model!
• EXAMPLE

28
Simulation

• XOR problem
• Other nonlinear problems
• Grids
• Short term forecasting
• M_C nonlinear function
• Henon function
• Etc

29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)
36
Applications

• Pattern recognition.
• Image processing.
• Prediction.
• Robotics.
• Control system.
• Etc.

37
Summary

• ANN technology represent a paradigm shift in real
  world problem solving techniques. With this
  technology, computers are now able to tackle
  problems whose underlying structure is not
  understood and now need not be understood.

38
PART-II Classroom LectureError Detection

• Introduction
• Error Detection Type
• Example-1 Simple Parity Check
• Example-2 Two Dimensional Parity Check
• Cyclic Redundancy Check (CRC)
• Example-3 Online CRC-Shift Register
• CheckSum
• Example-4 CheckSum
• Error Detection and correction Method

39
Introduction

• Data Communications
• Asynchronous One character (one byte of data) is
  transmitted
• Synchronous A block of characters is transmitted
• Note
  Data can be
  corrupted during transmission. For reliable
  communication, errors must be detected and
  corrected.
• There are two type of errors
• Single-bit Error
• Burst Error A block of characters is transmitted

40
Error Detection Type

• Single Bit Type
• In a single-bit error, only one bit in the data
  unit has changed
• Burst Type
• A burst error means that 2 or more bits in the
  data unit have changed

41
Example-1 Simple Parity Check

• Suppose the sender wants to send the word world.
  In ASCII the five characters are coded as
• 1110111 1101111 1110010 1101100 1100100
• The following shows the actual bits sent
• 11101110 11011110 11100100 11011000
  11001001
• Now suppose the word world is received by the
  receiver without being corrupted in transmission.
  
• 11101110 11011110 11100100 11011000
  11001001
• The receiver counts the 1s in each character and
  comes up with even numbers (6, 6, 4, 4, 4). The
  data are accepted. No error.
• Now suppose the word world is corrupted during
  transmission.
• 11111110 11011110 11101100 11011000
  11001001
• The receiver counts the 1s in each character and
  comes up with even and odd numbers
• (7, 6, 5, 4, 4). The receiver knows that the
  data are corrupted, discards them, and asks for
  retransmission.
• Simple parity check can detect all single-bit
  errors.

42
Example-2 Two Dimensional Parity Check (2-DPC)

• In 2-DPC, a block of bits is divided into rows
  and a redundant row of bits
• is added to the whole block.
• Suppose the following block is sent
• 11001111 10111011 01110010 01010011
  01010101
  
  
• But, some bits are corrupted.
• 11000001 00001011 01110010 01010011
  01010101
  
  
• When the receiver checks the parity bits, some of
  the bits do not follow the
• even-parity rule and the whole block is
  discarded.
• 11000001 00001011 01110010 01010011
  01010101

43
Cyclic Redundancy Check (CRC)

• The CRC is more reliable error detection method.
• The method is a mathematical division process.
• The entire string of bits in a block of data is
  divided by some preselected constant.
• The remainder is known as CRC.
• The CTC code is appended to the data stream.
• The receiver reads in the data plus the CRC bytes
  and performs division by same divisor.
• If the reminder will be zero if there were no
  errors in transmission.

44
Example-3 Online CRC-Shift Register

• Used JavaScript-Example

45
CheckSum

• The sender follows these steps
• The unit is divided into k sections, each of n
  bits.
• All sections are added.
• The sum is complemented and becomes the
  checksum.
• The checksum is sent with the data
• The receiver follows these steps
• The unit is divided into k sections, each of n
  bits.
• All sections are added to get the sum.
• The sum is complemented.
• If the result is zero, the data are accepted.

46
Example-4 CheckSum

• Suppose the following block of 16 bits is to be
  sent using a checksum of 8 bits.
• 10101001 00111001
• The numbers are added
• 10101001
• 00111001
  -------------
• Sum 11100010
• Checksum 00011101 complemented
• The pattern sent is 10101001 00111001
  00011101

47
Example-4 Continue

• Now suppose the receiver receives the pattern
  sent in is
• 10101001 00111001 00011101
• When the receiver adds the three sections, it
  will get all 1s,
• which, after complementing, is all 0s and shows
  that there
• is no error.
• 10101001
• 00111001
• 00011101
• Sum 11111111
• Complement 00000000
• Means that the pattern is OK.

48
Error Detection Correction

• Hamming Code Method
• This method of coding can detect and correct
  the error. For more information visit
• http//en.wikipedia.org/wiki/Hamming5FcodeHammi
  ng_codes

49

• Thank You

50
JAVA/APPLET Over Internet

• What is Applet?
• APPLET is an important feature of Java
  programming language that they can be easily
  accessible over the Internet using WWW browser
  such as Internet Explorer or others.
• How does Applet work?
• Write the Applet on the server side at the
• LOCATION IPaddress\WWW\AppletFilename
• Link the AppletFilename.class with index.html
  page

51
Example NameOfApplet.htm

• ltHTMLgt
• ltHEADgt
• lttitlegt welcome Java Applet lt/titlegt
• lt/HEADgt
• ltBODYgt
• ltPgtnbsplt/Pgtlt!-- Insert HTML here --gt
• ltappletcodeAppletProgramName.classwidth320h
  eight200
• lt/appletgtlt/BODYgtlt/HTMLgt