Altınbaş University Graduate Education Institute

Instructors: Asist. Prof. Dr. Aytuğ BOYACI

E-Mail: aytug.boyaci@altinbas.edu.tr
Web: www.aytugboyaci.com
Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 2
 Input: Output:
Data Artificial Action
Sensors Intelligence Movement
Images Text

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 3

1.An “AI”, or Artificial Intelligence is considered as an
intelligent machine made by human.
2.Most of time people think of it as a intelligent computer
program.
3.Ability to achieve goals by solving problems
4.Ability to correspond to the intellectual mechanism to the
task.
5.Minics cognitive functions such as learning, problem solving,
Planning.

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 4

1. Artificial Narrow Intelligence(ANI):

“weak AI”, specialized in one area. AI for chess game or go game

“strong Ai”, Human level AI where it is smart as human in many different

2. Artificial General Intelligence(AGI):
fields. Able to perform task like human. Currently no mature AGI, but
many companies and Orgs are working on this.

“The top”, far smarter than human in performing every task, trillions of
3. Artificial Superintelligence(ASI):
times smarter. Currently know only in Sci-Fi, and are mostly target by
people and media as a treat.

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 5

 First Computer invented: ENIAC- completed in
1946
 Alan Turing’s Paper “Computing Machinery
and Intelligence”, 1950
 Later The Turing test, Feigenbaum test
 The term “artificial Intelligence” , by John
McCarthy in 1956
 Loebner Prize-1990, 100,000 cash reward for
a system that can pass the Turing test, most
human like

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 6

 There has been several AI winters in the course of History
 Collapse of Lisp machine market in 1987, the fall of expert system
 In the late 90s and the early 2000s, AI was used in more
areas such as logistics, data mining and medical diagnosis.
 1997 Deep Blue from IBM defeated world chess champion
Gary Kasparov.
 In 2015 AlphaGo beat the world class Go player Lee Sedol
with 4 out of 5 wins. Proving proves on what AI can do in
one of human’s most complex game.

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 7

 Game Playing: Game AI plays an important role in the video
games you played. If you ever play single player mode in
games the NPCs are all game AI.
 Speech Recognition: Instruct machine to perform a certain
task by speaking. Needs a good understanding of the
language, and the mechanism on how to perform task.
 Natural Language Processing: Machine needs to understand
the meaning behind the text, with word prashing, grammar,
context etc. And output text that a human can
understanding.

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 8

 Computer Vision: machine takes input from camera,
process the data from the 3D world and then perform
a certain task with that data.
 Automation: Combine with other applications,
automation with AI can free up labor and have
enormous impact on the society today.
 Customer Service: Being about to serve human on
telephone and online, and solve the problem them
have.
Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 9
  Artificial Superintelligence: An intellect that is much smarter than
the best human brains in practically every field, including scientific
creativity, general wisdom and social skills.

 Artificial General Intelligence: A machine with the ability to apply

intelligence to any problem, rather than just one specific problem
(human-level intelligence)

 Artificial Narrow Intelligence: Machine intelligence that equals or exceeds

human intelligence or efficiency at a specific task

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 10

 Artificial
Intelligence

Machine Learning
Artifical
Machine Learning is a subset of
Artificial Intelligence

Deep Deep Learning uses

Deep Learning
Learnin neural networks to
is a subset of
g simulate human like
Machine Learning
decision making

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 11

 Machine Learning is a core sub area of
AI system that provides computers with
the ability to learn without explicitly
programing the steps.
 Recognizes pattern, and train the
algorithms with given input and output.
 Deep learning- a subfield of machine
learning with algorithms structures and
function like the brain called Artificial
neural network

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 12

 Easier to learn and use, more performance can be given
when large amount of data.
 Supervised - Give feedback on right on wrong, and adapt
to the feedback
 Unsupervised - Give no feedback, the machine should
learn the structure of the data to solve the task.
 Reinforcement - Isn’t giving feedback, only receive
feedback if it achieve the goal.

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 13

 Neural - cerebral cortex
Basic-level is called perceptron
Input layer - gives output
Hidden layer - Only activated by
node in previous layer.

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 14

 Type of Artificial Intelligence that provides computers with the ability to learn
without being explicitly programmed
 Various techniques can be used to for it learn make predictions based on data

Machine
Training Data Learning
Algorithm
Training
Prediction

“Live” Data Trained Model Prediction

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 15

  Supervised Learning: Learning with a labelled training set
 Example: email spam detector with training set of labelled emails

 Unsupervised Learning: Discovering patterns in unlabelled data

 Example: cluster similar documents based on the text content

 Reinforcement Learning: learning based on feedback or reward

 Example: learn to play chess by winning or losing

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 16

 Part of the machine learning field of learning representations of data. Expectably
effective at finding patterns
 Utilizes learning algorithms that derive meaning of day by using hierarchy of
multiple layers that mimic the neural network of our brain
 If you provide the system with tons of information it begins to understand it and
respond in useful ways.

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 17

INPUT A RESPONSE B APPLICATION
Picture Are there human faces? (0 or Photo tagging
1)
Loan Application Will they repay the loan? (0 Loan approvals
or 1)
Ad plus user information Will user click on ad? (0 or 1) Targeted online ads
Audio clip Transcript of audio clip Speech recognition
English Sentence French Sentence Language translation
Sensor from plane engine, etc Is it about to fail? Preventive maintenance
Car camera and other Position of other cars Self-driving cars
sensors

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 18

 Focuses on specific mundane task – coding invoices
 Built and maintained internally

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 19

 20+ years of coaching data
 Sold as software to be integrated with partner apps and devices

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 20

 Large dataset used for model training – over 30,000 soil types
 Partnership with university

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 21

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 22
Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 23
Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 24
  Uses Cameras to input dates to the computer
 Dynamic information collecting
 Data used to determine who you are
 Different algorithms were applied.
 75 percent
 https://www.captionbot.ai/

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 25

 Symbolic  Sub-symbolic
 Research on making AI  Research without specific
based on Human’s symbol representation of knowledge
representation of world  Nouvelle AI- produce AI for robot
 Logic based- no need for with enough intelligence to
human thought simulation control the body.
 cognitive simulation- Soar  Soft Computing - use Neural
cognitive architecture networks that use inexact
 Knowledge based solution to solve hard logic task
 Expert system- a first that can’t be solve using normal
successful AI software approach, tolerates imprecision.
 Need enormous about of
knowledge data for simple AI

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 26

 Intelligent agent - create an agent that can use different
approach, symbolic or subsymbolic, to solve the given
problem.
 Agent Architecture- A system uses the interaction
between multiple intelligent agent.
 Hierarchical control system- a control system where a set of
device or agent is arranged in Hierarchical tree.

Advanced Artificial Intelligence - Asst. Prof. Dr. .Aytuğ BOYACI 27