41

Neural Networks

Embed Size (px)

DESCRIPTION

basic concepts of neural networks

Citation preview

Page 1: Neural Networks
Page 2: Neural Networks

KNOWLEDGE-BASED KNOWLEDGE-BASED INFORMATION SYSTEMSINFORMATION SYSTEMS

Knowledge-based system is a Knowledge-based system is a program that acquires, represents program that acquires, represents and uses knowledge for a specific and uses knowledge for a specific purpose.purpose.

Consists of a knowledge-base and Consists of a knowledge-base and an inference engine.an inference engine.

Knowledge is stored in the Knowledge is stored in the knowledge-base while control knowledge-base while control strategies reside in the separate strategies reside in the separate inference engine.inference engine.

Page 3: Neural Networks

KNOWLEDGE-BASED INFO KNOWLEDGE-BASED INFO SYSTEMSYSTEM

Page 4: Neural Networks

WHAT ARE NEURAL NETWORKS ?WHAT ARE NEURAL NETWORKS ?

Artificial Neural Network (ANN) :- an information Artificial Neural Network (ANN) :- an information processing paradigm inspired by the HUMAN processing paradigm inspired by the HUMAN nervous system.nervous system.

Composed of large number of highly interconnected Composed of large number of highly interconnected processing elements (neurons).processing elements (neurons).

ANNs, like people, learn by example.ANNs, like people, learn by example.

An ANN is configured for a specific application, like An ANN is configured for a specific application, like pattern recognition or data classification, through pattern recognition or data classification, through learning.learning.

Learning in biological systems involves synaptic Learning in biological systems involves synaptic connections between neurons. This is true of ANNs connections between neurons. This is true of ANNs as well. as well.

Page 5: Neural Networks

Why use neural networks ?Why use neural networks ?

Knowledge acquisition under noise and Knowledge acquisition under noise and uncertainty.uncertainty.

Flexible knowledge representation.Flexible knowledge representation. Efficient knowledge processing.Efficient knowledge processing. Fault Tolerance .Fault Tolerance . They have learning capability.They have learning capability.

Page 6: Neural Networks

Neural networks versus Neural networks versus conventional computers conventional computers

ANNANN Learning approachLearning approach Not programmed Not programmed

for specific tasksfor specific tasks Used in decision Used in decision

makingmaking Operation is Operation is

unpredictableunpredictable

COMPUTERSCOMPUTERS Algorithmic approachAlgorithmic approach They are necessarily They are necessarily

programmedprogrammed Work on predefined Work on predefined

set of instructionsset of instructions Operations are Operations are

predictablepredictable

Page 7: Neural Networks

How does human brain learns ?

Brain ,made up of large no. of neurons. Each neuron connects to thousands of neurons,

communicates by electrochemical signals. Signals coming are received via SYNAPSES,

located at the end of DENDRITES. A neuron sum up the inputs, and if threshold

value is reached then it generates a voltage and o/p signal, along the AXON.

Page 8: Neural Networks

FIGURE SHOWING NEURON

Page 9: Neural Networks

SYNAPSE

Page 10: Neural Networks

-:THE ARTIFICIAL NEURON:-

Electronically modeled biological neuron.

Has many inputs and one output. Has 2 modes -training mode & using

mode. Training mode - neuron is trained to

fire (or not), for particular input patterns.

Page 11: Neural Networks

-:THE ARTIFICIAL NEURON :-

Using mode - when a taught input pattern is detected at input, its associated output becomes current output .

If input pattern does not belong in taught list, firing rule is used.

Page 12: Neural Networks

# ARTIFICIAL NEURON #

Page 13: Neural Networks

<((FIRING RULE))>

• Firing rule calculates whether neuron should fire for an input pattern or not.

• relates to all the input patterns, seen or unseen.

• The rule states :-• Take collection of training patterns for

node, some that cause it to fire (the 1-taught set of patterns) and others which prevent it from firing (the 0-taught set).

Page 14: Neural Networks

<((FIRING RULE))>

• Then, the patterns not in collection cause node to fire if, they are more similar to patterns in the 1-taught set, than with patterns in the 0-taught set. If there is a tie, then pattern remains in undefined state.

Page 15: Neural Networks

<((FIRING RULE))>

Example :• a 3-input neuron is taught to output

1 when the input (X1,X2 and X3) is 111 or 101 and to output 0 when the input is 000 or 001.

• Now, if we give 010,then the neuron will not fire, for 011 o/p is undefined.

Page 16: Neural Networks

~:“PATTERN RECOGNITION”:~

Pattern recognition is implemented by using neural networks.

During training, the network is trained to associate outputs with input patterns.

The n/w then identifies the input pattern and tries to output associated output pattern.

Page 17: Neural Networks

~:“PATTERN RECOGNITION”:~

The power of neural networks comes to life when a pattern that has no output associated with it, is given as an input.

In this case, the network gives the output that corresponds to a taught input pattern that is least different from the given pattern.

Example :- recognition of alphabets, symbols etc.

Page 18: Neural Networks

Here the inputs are weighted inputs.Effect of an input on decision making is

directly proportional to the weight of that input.

Weight is a floating pt. number, can be +ve or –ve.

As each input enters the nucleus it is multiplied by its weight .

Page 19: Neural Networks

Neuron then sums these new input values which gives us the activation .

If activation is greater than threshold value, the neuron outputs a signal, else zero output.

This is typically called a step function .

Page 20: Neural Networks

# ARTIFICIAL NEURON WITH #WEIGHTED INPUTS

Page 21: Neural Networks

In mathematical terms, neuron fires if and only if,

X1W1 + X2W2 + X3W3 + ... > T The MCP neuron has the ability to adapt to a

particular situation by, changing its weights and/or threshold.

Various algorithms exist that cause the neuron to 'adapt'; the most used ones are the Delta rule and the back error propagation.

Page 22: Neural Networks

Architecture Of Neural Architecture Of Neural Networks Networks

FEED –FORWARD NETWORKS :-FEED –FORWARD NETWORKS :-1.1. allow signals to travel one way only ; allow signals to travel one way only ;

from input to outputfrom input to output . .

2.2. no feedback (loops) i.e. the output of any no feedback (loops) i.e. the output of any layer does not affect that same layer. layer does not affect that same layer.

3.3. Feed-forward ANNs tend to be straight Feed-forward ANNs tend to be straight forward networks that associate inputs forward networks that associate inputs with outputs. with outputs.

4.4. extensively used in pattern recognitionextensively used in pattern recognition . .

Page 23: Neural Networks

FEED-FORWARD FEED-FORWARD NETWORKNETWORK

Page 24: Neural Networks

Architecture Of Neural Architecture Of Neural Networks Networks

FEEDBACK NETWORKS :-FEEDBACK NETWORKS :-1.1. can have signals traveling in both can have signals traveling in both

directions by introducing loops in the directions by introducing loops in the network. network.

2.2. Feedback networks are dynamic; their Feedback networks are dynamic; their 'state' is changing continuously until 'state' is changing continuously until they reach an equilibrium point. they reach an equilibrium point.

3.3. They remain at the equilibrium point They remain at the equilibrium point until the input changes and a new until the input changes and a new equilibrium needs to be found .equilibrium needs to be found .

4.4. also referred to as interactive or also referred to as interactive or recurrent .recurrent .

Page 25: Neural Networks

FEEDBACK NETWORKFEEDBACK NETWORK

Page 26: Neural Networks

Architecture Of Neural Architecture Of Neural NetworksNetworks

Network layersNetwork layers :- :-

Artificial neural network mostly consists of Artificial neural network mostly consists of three groups, or layers, three groups, or layers,

Input Layer - activity of input units represents Input Layer - activity of input units represents raw information that is fed into the network.raw information that is fed into the network.

Hidden Layer - activity of each hidden unit is Hidden Layer - activity of each hidden unit is determined by activities of input units and determined by activities of input units and weights on connections between input and weights on connections between input and hidden units.hidden units.

Output Layer - behavior of the output units Output Layer - behavior of the output units depends on the activity of the hidden units and depends on the activity of the hidden units and the weights between the hidden and output the weights between the hidden and output units.units.

Page 27: Neural Networks

Layers Of Neural Layers Of Neural NetworkNetwork

Page 28: Neural Networks

Architecture Of Neural Architecture Of Neural NetworksNetworks

Perceptrons :-Perceptrons :-1.1. It is an MCP model It is an MCP model

with some with some additional, fixed, additional, fixed, pre-processing .pre-processing .

2.2. Units A1, A2, Aj , Ap Units A1, A2, Aj , Ap are called are called association units and association units and theythey extract specific, extract specific, localized features from localized features from input images. input images.

3.3. mimic the basicmimic the basic idea idea behind the human behind the human visual system. visual system.

Page 29: Neural Networks

Most common methods used Most common methods used are :-are :-

1.1. Supervised LearningSupervised Learning

2.2. Unsupervised LearningUnsupervised Learning

Page 30: Neural Networks

Incorporates an external teacherIncorporates an external teacher Each output unit is told what its Each output unit is told what its

desired response to input signals desired response to input signals should be.should be.

During learning process global During learning process global information is required.information is required.

Supervised learning includes Supervised learning includes error- correction and error- correction and reinforcement learning. reinforcement learning.

Page 31: Neural Networks

There is problem of error There is problem of error convergence i.e. minimization of convergence i.e. minimization of error between desired and error between desired and computed values .computed values .

The aim is to determine a set of The aim is to determine a set of weights which minimizes the weights which minimizes the error.error.

A well-known method is least A well-known method is least mean square (LMS) convergence. mean square (LMS) convergence.

Page 32: Neural Networks

Uses no external teacher, based Uses no external teacher, based upon only local information. upon only local information.

Also referred to as self-organizing, in Also referred to as self-organizing, in sense that it self-organizes data sense that it self-organizes data presented to network, detects their presented to network, detects their emergent collective properties. emergent collective properties.

Methods of unsupervised learning Methods of unsupervised learning are Hebbian and competitive are Hebbian and competitive learning .learning .

Page 33: Neural Networks

Learning consists of two phases- Learning consists of two phases- training phase and operation phase.training phase and operation phase.

We say that neural network learns off-We say that neural network learns off-line if learning phase and operation line if learning phase and operation phase are distinct.phase are distinct.

A neural network learns on-line if it A neural network learns on-line if it learns and operates at the same time.learns and operates at the same time.

Usually, supervised learning is Usually, supervised learning is performed off-line, whereas performed off-line, whereas unsupervised learning is performed on-unsupervised learning is performed on-line.line.

Page 34: Neural Networks

• Character RecognitionCharacter Recognition

• Since neural networks are best at Since neural networks are best at identifying patterns or trends in data, identifying patterns or trends in data, they are well suited for prediction or they are well suited for prediction or forecasting needs including:forecasting needs including:

1.1. sales forecasting sales forecasting

2.2. industrial process control industrial process control

3.3. data validationdata validation 4.4. risk management risk management

Page 35: Neural Networks

• Neural networks are also used forNeural networks are also used for1.1. Genetic pattern recognitionGenetic pattern recognition2.2. Drug discoveryDrug discovery3.3. Flow Cytometric Analysis of LeukemiaFlow Cytometric Analysis of Leukemia

• Also used in field of Robotics , Also used in field of Robotics , Facial Animation, Lip Reading , Facial Animation, Lip Reading , Event Prediction Event Prediction

and many more fields. and many more fields.

Page 36: Neural Networks

Learning modeLearning mode Prediction modePrediction mode

Page 37: Neural Networks

Shape of LipsShape of Lips Lip readingLip reading

Page 38: Neural Networks

• The computing world has a lot to gain from neural networks.

• Their ability to learn by example makes them very flexible and powerful.

• Further, there is no need to devise an algorithm in order to perform a specific task; i.e. there is no need to understand the internal mechanisms of that task.

• They are also very well suited for real time systems because of their fast response and computational times which are due to their parallel architecture.

Page 39: Neural Networks

• Neural networks also contribute to other areas of research such as neurology and psychology.

• Finally, I would like to state that even though neural networks have a huge potential we will only get the best of them when they are integrated with computing, AI and related subjects.

Page 40: Neural Networks

en.wikipedia.orgen.wikipedia.org www.sciencedirect.comwww.sciencedirect.com www.learnartificialneuralnetworks.comwww.learnartificialneuralnetworks.com www.doc.ic.ac.ukwww.doc.ic.ac.uk www.statsoft.comwww.statsoft.com NEURAL NETWORKS IN COMPUTER NEURAL NETWORKS IN COMPUTER

INTELLIGENCE ,by – LiMin FuINTELLIGENCE ,by – LiMin Fu

Page 41: Neural Networks