neuro network1

18
ARTIFICIAL NEURAL NETWORKS PRESENTED BY : KOMAL SHARMA B.Tech(IT)-III yr

Upload: komal-sharma

Post on 15-Apr-2017

366 views

Category:

Technology


0 download

TRANSCRIPT

Page 1: Neuro network1

ARTIFICIAL NEURALNETWORKS

PRESENTED BY: KOMAL SHARMA B.Tech(IT)-III yr ROLL No.-7199

Page 2: Neuro network1

CONTENTS…

The following points are covered in this presentation: Introduction History Inspiration from biological neurons Architecture of neural network Working of artificial neurons Characteristics Applications Advantages and disadvantages Future scope conclusion

Page 3: Neuro network1

INTRODUCTION An interconnected group of nodes , akin to the vast network of

neurons in a brain A computational model inspired in the natural neuron An attempt at modelling the information processing capabilities

of nervous systems. An biological approach to Artificial Intelligence Process information by their dynamic state response to external

inputs

A basic artificial neuron network

Page 4: Neuro network1

HISTORY….1943 Warren McCulloch &

Walter PitsComputational model for neural networks

1950 Possible to simulate a hypothetical neural network

1958 Frank Rosenblatt Formation of perceptron

1959 Bernard Widrow & Marcian Hoff

MADALINE-first neural network

1962 Neural research went down drastically and was left behind

1972 Kohonen &Anderson Similar network independently

1975 First multilayered network

1982 John Hopfield Renewed interest1990s-present Continuous advances in various fields

Page 5: Neuro network1

INSPIRATION FROM BIOLOGICAL NEURONS

Examinations of humans’ CNS inspired the concept of artificial neural networks

Animals react adaptively to changes in their external and internal environment-use their nervous system to perform these behavior

An appropriate/simulation of the nervous system should be able to produce similar responses and behaviors in artificial systems.

A biological neuron An artificial neuron

Page 6: Neuro network1

ARCHITECTURE…. NETWORK LAYERS a) Input Layer b) Hidden Layer c) Output Layer RECURRENT STRUCTURE-Feedback Networks NON-RECURRENT STRUCTURE-Feedforward Networks

Network layers structure

Page 7: Neuro network1

FROM HUMAN NEURONS TO ARTIFICIAL NEURONS…..

Try to deduce the essential features of neurons and their interconnections

A computer is programmed to simulate features

Incomplete knowledge of neurons and limited computing power result into..

Necessarily gross idealizations of real networks of neurons

The neuron model A basic artificial neuron

Page 8: Neuro network1

WORKING…..of ANNs

Perceptron- artificial neuron

Electrical signals as numerical values

A network of neurons is formed

Page 9: Neuro network1

WORKING…..of ANNs

Principle used… FIRING RULES Determine how one calculates whether one should fire for any input pattern Some sets which cause it to fire have 1-taught set of patterns and others

which do not have 0-taught set. Accounts for high flexibility For example: Suppose there is 3-input neuron which is taught to produce output 1 when the input is 111 or 101 and outputs 0 when the input is 000 or 001.

Page 10: Neuro network1

CHARACTERISTICS… Parallel Processing Ability

Distributed Memory

Fault Tolerance Ability

Collective Solution

Learning Ability

Page 11: Neuro network1

APPLICATIONS….. Pattern Recognition

Character Recognition

Prediction of stock price index

Neural networks in Medicine

Travelling Salesman’s Problem

Airline security control

Page 12: Neuro network1

AN EXAMPLE…stock market predictionTraining data

This month’s stock priceUnadjusted retail salesindustrial production index

Govt. receipts

Govt. expenditures

Gold price

Dollar value

Input layer Hidden layer

Next month’s stock price

Output layer

Page 13: Neuro network1

ADVANTAGES….

Perform tasks that a linear program can not do.

A neural network learns and does not need to be reprogrammed

It can be implemented in any application

No algorithm is required. They learn by examples.

Page 14: Neuro network1

DISADVANTAGES… Training is needed to operate neural network.

Emulation is needed because architecture of neural network is different from the architecture of the microprocessors.

High processing time is required for large neural networks.

Not a general purpose problem solver.

No structured methodology.

Page 15: Neuro network1

RECENT ADVANCES &FUTURE APPLICATIONS…

Integration of fuzzy logic into neural networks

Pulsed Neural Networks

Improvement of existing technology

Common usage of self-driving cars

Robots that can see, feel or predict the world around them

Page 16: Neuro network1

CONCLUSION…Computing world to gain a lot from neural networks

Have a very promising future due to its flexibility

Possibility that some day “conscious” networks might be produced

Despite having a huge potential, these are best used only when they are integrated with computing, AI, fuzzy logic and related subjects

Page 17: Neuro network1

REFERENCES… https://en.wikipedia.org/wiki/Artificialneuralnetwork www.psych.utoronto.ca/users/reingold/courses/ai/cache/neural2.

html www.doc.ic.ac.uk/~nd/surprise96/journal/vol4/cs11/report.html https://datajobs.com/data-science.../Neural-net[carlos-Gershens

on].pdf www.cse.unr.edu/~bebis/Mathematical/NNs/lecture.pdf www.softcomputing.net/annchapter.pdf pages.cs.wisc.edu/~bolo/shipyard/neural/local.html

Page 18: Neuro network1