ARTIFICIAL NEURAL

NETS
Introduction into the artificial intelligence - 1.

Zuzana Komínková Oplatková

© 2013-2017

!1
ARTIFICIAL NEURAL NETS

where do we put artificial neural nets?

Artificial intelligence
is the field of the informatics which deals with the creation of machines
showing signs of intelligent behavior.
1955 - John McCarthy proposed the using of the term artificial
intelligence

!2
 INTRODUCTION INTO AI 1.
humans have tried to create a machine or a system which
would make work easier and would be also “intelligent”
effort to create such systems appears usually in movies firstly,
in sci-fi and sometimes come to truth in reality

!3
 INTRODUCTION INTO AI 2.
autonomous systems - a car does not need any driver,
it is oriented by means of computer systems in the
environment (detection of the surroundings) like -
radar, GPS, lidar (similar to radar but with laser,
amount of points -> interpolation into the 3D digital
model of the surface, buildings, etc.), computer vision

!4
 INTELLIGENCE ?
what intelligent behavior means is still the subject of discussion

What is intelligence?
„If a machine reacts on the stimulus of the human partner so that a human is not able to
recognize if he/she deals with the machine or the person by means of the terminal, it is
considered as intelligent.“ – Turing

Marvin Minsky claims, that „… artificial intelligence is science about the creation of
machines or systems which will use such a procedure, which - if a human would do it -
will be considered as an expression of its intelligence. “ (1967)

E. Richová believes that „… artificial intelligence deals with solving of tasks by means of
computers which humans handle better currently.“ (1991)

AI is designation of artificially created event which reminds convincingly a natural

phenomenon of human intelligence.

!5
 TURING TEST 1.
A. Turing (called the test “imitation test”) tried to create a
test which will answer a question “Can machines think?”
He wanted to transfer the philosophical questions into exact
research level.
Test should check out if the system of the artificial
intelligence behaves really intelligently
We can say that a thinking machine is a machine (computer)
if we will not recognize its behavior from the human
behavior.
definition and testing of the intelligence is complicated
therefore comparison with a human
!6
 TURING TEST 2
we put a tester, object of investigation (e.g. a computer with
a software) and another person into separated rooms
tester gives questions and then randomly it is selected if an
answer is given by a computer or by a human
answers are given back to a tester in a neutral way (e.g.
printed)
if a tester does not recognize who is “talking with”, then the
system is considered as intelligent, i.e. fulfill the Touring test

!7
 TURING TEST 3.
Part of the dialogue between ELIZA (E) and patient (P): (1966)


P: Men are all alike.

E: In what way?
P: They're always bugging us about something or other.
E: Can you think of a specific example?
P: Well, my boyfriend made me come here.
E: Your boyfriend made you come here.
P: He says I'm depressed much of the time.
E: I am sorry to hear that you are depressed.
P: It's true, I am unhappy.
E: Do you think coming here will help you not to be unhappy?
P: I need some help; that much seems certain.
E: What would it mean to you if you got some help?
P: Perhaps I could learn to get along with my mother.
E: Tell me more about your family.
P: My mother takes care of me.
E: Who else in your family takes care of you?
P: My father.
E: Your father.
P: You are like my father in some ways.
E: What resemblance do you see? ...

!8
 TURING TEST 4
Program ELIZA we classify as a chatterbot today
it fulfills the Touring test only partly
non intelligent program is able to communicate seemingly
intelligently
but still an award of 100 000 USD has not been given yet
for the program which will be completely unrecognizable
from a human
Loebner's award - each year competition for software
simulating of a human communication since 1990
!9
CHINESE ROOM ARGUMENT 1.

a thought experiment
John Searl - year 1980
it illustrates an imperfection of Touring test
the ability of meaningfully answering to given questions is
not enough to prove the ability of understanding, which is
the most important what we expect from the strong AI

!10
CHINESE ROOM ARGUMENT 2.

closed room full of Chinese texts, where there are meaningful sentences of this
language

we put a human inside who is not able to speak Chinese at all but he has a
knowledge how to find answers to questions.

we will give written questions to this person (as in Turing test). This person is able
to find enough materials in this library to be able to find a meaningful answer to a
given question and give it outside.

The outside questioner could think that a person inside is able to understand
Chinese without problems even it is not so in reality. A person inside works only
mechanically with unknown symbols for him so also a not thinking machine
would be able to do the same.

!11
ARTIFICIAL NEURAL NETS
where do artificial neural nets belong?

Artificial intelligence
soft computing
neural nets, learning algorithms (Support Vector Machines)
evolutionary algorithms, symbolic regression, metaheuristics system
machine learning
pattern recognition
fuzzy logic
self-adaptive systems
chaos theory
datamining
expert systems
Bayesian networks !12
 SOFT VS. HARD COMPUTING
Inverted pendulum on a automatically controlled cart

2 approaches

we will study physics, we describe a given process by means of partial differential

equations, we implement a numerical algorithms into a computer which calculates
approximate result and then we obtain a controlling function for a given cart - Hard-
Computing

in the second case, no matter which physical laws control the balancing of the
pendulum. We will observe outer knowledge which does not describe reasons
of behavior but the behavior itself. 

This knowledge will be given e.g. in the form of IF - THEN rules: If a pendulum is
deflected slightly to the right, the cart will be moved also to the right. There are ruls
describing a logical dependence between deviation and reaction - Soft-Computing
!13
 SOFT COMPUTING
soft - tolerant to inaccuracy, partly true, approximative
a model was a human brain
the purpose to achieve a robustness and low expenses for solution
term was established by Dr. Lofti A. Zadeh - 1973
it covers mainly - fuzzy (logic, modeling, decision and data analysis),
later also artificial neural nets and genetic computation techniques
(today rather evolutionary)
definition is still opened and still is “adding, completing”, it belongs
here also hybridization methods, combination of methods
it belongs also as subregion of AI
more in 5. year - course Soft computing
!14
 REGIONS OF USAGE OF AI
natural and artificial thinking, cognition as a philosophical
problem,
multi agent systems,
solving of tasks, proving of clauses and usage of knowledge,
mathematical theory of AI
representation of knowledge and modeling, including a theory of complexity,
recognition and learning, applied AI and expert systems.
planning,
analysis, synthesis and recognition of speech,
processing of natural language,
computer vision, pattern recognition
knowledge engineering,
distributed artificial intelligence,
programming means of AI,
evolutionary methods, self reproduction and artificial life
!15
 ARTIFICIAL LIFE
NORNS
creatures - a game, but also scientific experiments with
fighter
they showed intelligent behaviour - reproductive
organs were given to them but no “reproduction” was
programmed and even though an offspring was
created
fighter - found a possibility of rotation

!16
 NEURAL NET 1.

Artificial neural nets manage to do various things but

it is also necessary to pay attention what we want to
achieve from them
NN might show bad or unacceptable behaviour
because they carry e.g. information about objects
which do not exist in reality

!17
NEURAL NET 2.

!18
 NEURAL NET 3.
Effect “memory switch”
the brain is able to find the image on
Object rectangle the basis of experiences
does not exist in once you see and you will never forget!
the image

!19
 NEURAL NET 4.
Communication by means of the voice has not been
ideal yet
It is necessary to develop not only neural nets but also
other scientific fields - signal processing, acoustics,
phonetics, speech processing, pattern recognition,
linguistics...

!20
 NS - APPLICATION 1.

writing recognition

!21
 NS - APPLICATION 2.
prediction, eruption on the sun, time series on the stock
market...

!22
 COMPUTER VISION
digital image acquisition
adjustment, decomposition on objects and their
description
classification of objects

!23
 COMPUTER VISION 2.
cooperation of computer graphics and artificial
intelligence
cost - saving description
usage during the graphical object search - e.g. Google

stork_carries_child.jpg whatever.jpg
!24
 COMPUTER VISION 3.
is similar to study of biological vision
can be used for movie analysis too
analysis of medicine data (search for tumour objects, objects
marking different illnesses - e.g. in an eye, analysis of EEG,
etc.)
analysis of the environment for the autonomous system
movement, robots, etc.
it is a interdisciplinary field - there is no clear task of
computer vision - it belongs to AI or to informatics generally
neural nets are used often for the computer vision
applications
!25
 INTERNET, VIRUSES
optimization of Internet running, computer usage of
sources (data storage, cpu use….)
detection of viruses - via neural nets - Symantec, AVG
(AVG Technologies)

!26
EVOLUTIONARY ALGORITHMS

evolutionary algorithms - winter term - 5th year.

work as optimization algorithms

!27
 EVOLUTIONARY ALGORITHMS 2.

evolutionary Definition of
parameters
Generation of
1st random
proces population
Cost value of
parents
terminology Change of old
population by new
taken from one

nature Selection of best Selection of

individuals parents

the aim is to
find the Cost value of
New offspring
optimal offspring

solution of Mutation of

the given task offspring

!28
EA - APPLICATION

!29
 EVOL. ALG. A NEUR. NET

methods of
AI is possible
to combine
EA can adjust
also a neural
net

!30
 AI AND SOME VISIONS

AI and amusement… (games, music…)

AI and Internet of Things

AI and future

!31
 AI AND AMUSEMENT

evolutionary music

games with AI inside (since

1978…)

!32
AI AND INTERNET OF THINGS
Internet of Things
internetworking physical devices (connected devices),
buildings and other items - embedded with electronics,
software, sensors, actuators, and network connectivity
that enable these objects to collect and exchange data
the infrastructure of the information society
Each thing is uniquely identifiable through its embedded
computing system but is able to interoperate within the
existing Internet infrastructure.
AI is must have to be competitive on the market
“Hey, fridge. When you will order the milk?” or “Buddy,
wash them in economy mode.”
!33
 AI AND FUTURE
repair of microstructures,
elimination of ecological
catastrophes

exploration of unknown
planets

different forms - not only a

human but also a bird or
fish

!34
 AI AND FUTURE 2.
implants for improvement of health (hearing...)?
or Cyborgs?

!35
 FIRST CYBORG

prof. Kevin Warwick

he let implant RFID
chips into his arm
!36
 KEVIN WARVICK
when moving in the building - implanted silicone chip (24th
August 1998) - allowed on/off lights, door controls, heating
second project - 100electrode array - 14th March 2002
allows sending and receiving signals between an implant,
nervous system and a computer - controlling of a remote
hand, wheelchair
if successful (it was) also his Czech wife obtains (ed) a
chip, and they will researched if it is possible to send
signals of emotions from one person to another one and
possibly react back
https://www.youtube.com/watch?v=Z8HeFNJjuj0
https://www.youtube.com/watch?v=cx45D9aWEeY
!37