Intelligent Information Systems

cracklegulleyΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

86 εμφανίσεις

page

1

M. Muraszkiewicz


Intelligent Information

Systems


Prof. M. Muraszkiewicz






Institute

of Information
and Book Studies

Warsaw University


mietek@n
-
s.pl

page

2

M. Muraszkiewicz

Neural Nets



Module 10

page

3

M. Muraszkiewicz

1.
Background

2.
Historical Note

3.
Definition

4.
Properties and Applications

Table of Contents

page

4

M. Muraszkiewicz






Background

page

5

M. Muraszkiewicz

Two Tracks in AI


Analytical, symbolic



Invented by researchers


(
inspired by logics and math



J. von Neumann)
.



Naturalistic





Based on solutions worked
out by “mother nature”
through evolution




(
inspired by psychology, neurology, biology,
evolution



K. Darwin, ...)
.


page

6

M. Muraszkiewicz

About the Human Brain



If the human brain
were so simple that
we could understand
it, we would be so
simple that we
couldn’t.



Emmerson M. Pough

page

7

M. Muraszkiewicz

Parameters



volume
:
~

1400 cm3,


weight
:
~

1,5 kG,


surface
:
~

2000 cm2
(the surface of a sphere of

the same volume is
~

600 cm2)
,


~

10
10

neurons
,


10
12

glia cells
,


number of connections

-

~

10
15

average length from
0,01 mm
to

1m.



Neurons receive and send impulses whose frequency is


1
-

100 Hz,
duration

1
-

2 ms,
voltage

100 mV
and

speed of
propagation

1
-

100 m/s.


Speed of brain



10
18

operations
/s (
parallel processing
).


Informational capacity of senses’ channels
:

--

vision

-

100 Mb/s,

--

touch

-

1 Mb/s,

--

audition

-

15 Kb/s,

--

smell

-

1 Kb/s,

--

taste

-

100 b/s. (
source

R. Tadeusiewicz, „Sieci neuronowe”).


page

8

M. Muraszkiewicz






Historical Note

page

9

M. Muraszkiewicz

Difficult History



W. McCulloch, W. Pitts



first mathematical model
of a neuron

(1943),



D. Hebb



the rule that

determines the change in
the weight connection,



F. Rosenblatt
’s
Perceptron

(1957)
, a two
-
layer
network,

for recognizing alphanumerical
characters
,



B. Widrow, M. Hoff



ADALINE



M. Minsky (1969)


proved limits of simple neural
nets which weakened research in the 70’ies
,



J. Hopfield
’s Net

with a feedback

(1982),



Works by

J. Andersona (1988)


neural nets’

comeback
".

Warren McCulloch

1898
-
1969

page

10

M. Muraszkiewicz






Definition

page

11

M. Muraszkiewicz

Intuitive

Definition

“A neural network is a set of simple
processors (“neurons”) connected in a certain
way
.

A neuron can have many inputs
(synapses) with which weights can be
associated.


The value of weights can be changed during
the operation of a network to produce the
desired data flow within it what makes the
network and adaptive device.


Topology of the network and the values of
weights determine the program executed on
the network
.

page

12

M. Muraszkiewicz

Definition from Wikipedia


An artificial neural network (ANN),
often just called a "neural network"
(NN), is an interconnected group of
artificial neurons that uses a
mathematical model or computational
model for information processing
based on a connectionist approach to
computation




http://en.wikipedia.org/wiki/Artificial_neural_network

page

13

M. Muraszkiewicz

Types of Nets

The neurons learn in an iterative
way.



By adding an error detector and a
feature to change weights simple
nets become to new models such
as
ADALINE

(ADAptive LINear
Element).

F
eedforward

Linear

One
-
layer

other

With a
feedback

(Hopfield)

Non
-
linear

Multi
-
layer

page

14

M. Muraszkiewicz






Properties

and Applications

page

15

M. Muraszkiewicz

Main Properties

Advantages



adaptiveness and self
-
organization



parallel processing
,



learning (supervised and unsupervised)



fault tolerance


Disadvantages



non
-
explicability



slow

page

16

M. Muraszkiewicz

Type of Applications


prediction


optimization


classification


pattern and sequence recognition


data analysis and association
,


filtering


...

page

17

M. Muraszkiewicz

Examples of Applications


Diagnostics of electronic
devices



Psychiatric research



Stock exchange predictions



Sales predictions



Search for oil fields



Interpretation of biological
research



Prices prediction



Analysis of medical data



Planning of machines
maintenance


Speech analysis



Planning of learning progress



Analysis of production problems



Trade activities optimization



Spectral analysis



Optimization of wastes utilization



Selection of row materials



Forensic support



Staff recruitment support



Industrial processes control



...

page

18

M. Muraszkiewicz

Readings


Haykin S., “
Neural Networks: A Comprehensive

Foundation


(3rd Edition)
,
Prentice Hall
, 2007.



Lawrence, J
., “
Introduction to Neural Networks

,

California Scientific Software Press
, 1994.



Royas R., “
Neural Networks: A Systematic Introduction
”,
Springer, 1996.




http://en.wikipedia.org/wiki/
Neural_networks

http://en.wikipedia.org/wiki/Artificial_neural_network

page

19

M. Muraszkiewicz