Introduction to

AI
TAE-HOON KIM
SLIDE 01
 Industrial Revolutions

Industrial Revolutions
01 02 03 04

Industry 1.0 Industry 2.0 Industry 3.0 Industry 4.0

First industrial revolution: Second industrial Third industrial revolution: Fourth industrial
transition from hand revolution: technological Digital revolution, which revolution: combine
power to machine power revolution, supported by made significant progress hardware, software, and
using steam and water extensive railroad network in communication biology with advances
power and telegraph technologies in communication and
connectivity
 Cyber-physical systems Cognitive computing
Integration of computation, Simulate human thought
networking, and physical processes in a computerized
processes. Control and monitor model using self-learning
the physical process with algorithms, which utilize data
feedback mining, pattern recognition and
natural language processing

IoT Cloud computing

System of interrelated identified On-demand availability of
computing devices that can computer system resources,
exchange data without such as storage and computing
requiring human-to-human or power, without direct active
human-to-computer interaction
Industry 4.0 management by the user
 Smart or Intelligent?
Smart Intelligent refers to the ability to
acquire knowledge
Factory
Smart refers to the ability to apply
previously acquired knowledge in
practical situations

Smart Factory is..

Vision of a production environment
in which production facilities and
logistics systems are organized
without human intervention.

The smart factory represents a leap

forward from traditional automation
to a fully connected and flexible
system
Examples of Smart Factory
Whirlpool Siemens
Application manufacturer MindSphere – cloud-based IoT open operating
Set itself a goal of eliminating the waste that its System (Platform as a Service (PaaS))
factories send to landfills by 2020.
Use an analytics platform across all of its It provides the access to this platform via open
facilities to reveal the amounts of waste they interfaces and use it for their own analysis.
generate, along with usage data about For example, the online monitoring of globally
electricity, water, and more. distributed machine tools, industrial robots, or
Retrieving local information, the company can industrial equipment.
check the sustainability performance of its It also allows customers to create digital models
plants worldwide. of their plants using real data
Company can measure how close any factory
is to the zero-waste milestone, no matter where
it may be in the world.
Examples of Smart Factory
FANUC KUKA
In 2016, collaborated with Cisco and Rockwell Invested in improvement of human-robot
Automation – FIELD (FANUC Intelligent Edge collaboration using AI
Link and Drive) – IIoT platform for
manufacturing. LBR iiwa – uses intelligent control technology
and high-performance sensors for HRC
(Human Robot Collaboration)
Partnered with NVIDIA to use Artificial
intelligence (AI)

FANUC uses deep reinforcement learning to

train the Industrial robot
 D i g i t a l Tw i n
Concept since 2002
Virtual model of a process or product

Forming a foundation for

connected product and services You can simply impress your audience and add a unique
zing and appeal to your Presentations. Get a modern
PowerPoint Presentation that is beautifully designed. I
hope and I believe that this Template will your Time,
Physical Bridget Digital Money and Reputation.
Image: https://government.economictimes.indiatimes.com
Artificial Intelligence
Self-Driving Car
Robots
Outline
Discuss about Artificial Intelligence
AI and Machine learning
What is AI?
Rational Decisions
Rational: Maximally achieving pre-defined goals
Rationality only concerns what decisions are made
◦ not the thought process behind them
Goals are expressed in terms of the Utility of outcomes
Being rational means Maximizing your expected utility
Maximize
Your
Expected
Utility
Maximize
Your
Expected
Utility
What about the
Brain?
▪ “Brains are to intelligence as wings
are to flight”
▪ Brains (human minds) are very good
at making rational decisions, but not
perfect
▪ Brains aren’t as modular as software,
so hard to reverse engineer!
▪ Lessons learned from the brain:
memory and simulation are key to
decision making
Designing Rational Agents
An agent is an entity that perceives
and acts.
A rational agent selects actions that
maximize its (expected) utility.
Characteristics of the percepts,
environment, and action space

Environment
Sensors
Percepts

Agent
dictate techniques for selecting rational ?

actions Actuators
Actions
Brief history of AI
Intelligence
Capacity for logic, understanding, self-awareness, emotional
knowledge, reasoning, planning, creativity, critical thinking, and
problem-solving
It should be able to
◦ Interact with the real world
◦ Perform Reasoning and planning
◦ Learn and adapt
What Can AI Do?
Play a decent game of Jeopardy?
Win against any human at chess?
Win against the best humans at Go?
Play a decent game of tennis?
Grab a particular cup and put it on a shelf?
Unload any dishwasher in any home?
Drive safely along the highway?
Drive safely along Calumet Avenue?
Buy a week's worth of groceries on the web?
Buy a week's worth of groceries at Whole Food Market?
Discover and prove a new mathematical theorem?
Perform a surgical operation?
Translate spoken Korean into spoken English in real time?
Write an intentionally funny story?
Deep Blue (IBM) – Game Agent
Classic Moment: May, '97: Deep Blue vs. Kasparov
◦ First match won against world champion
◦ “Intelligent creative” play
◦ 200 million board positions per second
◦ Humans understood 99.9 of Deep Blue's moves
◦ Can do about the same now with a PC cluster
1996: Kasparov Beats Deep Blue
“I could feel --- I could smell --- a new kind of intelligence across
the table.”
1997: Deep Blue Beats Kasparov
“Deep Blue hasn't proven anything.”
Search problem formulation
States – environment
Initial state – the agent starts in
Actions – available actions to the agent
Transition model – returns the state resulting from action
Goal states – where the agent stops
Action cost – numeric cost of applying action in given state
Example – Sokoban puzzle
States
◦ State description specifies the location of each of the tiles
Initial state
◦ Any state can be designated as the initial state
Actions
◦ Physically, we move tiles
◦ In this problem, possible move of empty box
◦ (L, R, U, D), (R, U, D), (L, U, D), (R, U), (R, D), (L, U), (L, D)

Transition model
◦ Maps a state and action to a resulting state
Goal state
Action cost → Each action cost is 1
State Space Graphs
State space graph: A mathematical
representation of a search problem
◦ Nodes are (abstracted) world configurations
◦ Arcs represent successors (action results)
◦ The goal test is a set of goal nodes (maybe only one)

In a state space graph, each state occurs only

once!

We can rarely build this full graph in memory

(it’s too big), but it’s a useful idea
Uncertainty in the real world
Your action does not result in deterministic state anymore.
There is an uncertainty.
The result of your action has a randomness

S’1
Random

S action a

S’2
Markov Decision Processes
An MDP is defined by: M = (S,A,T,R)
◦ A set of states s  S
◦ A set of actions a  A
◦ A transition function T(s, a, s’)
◦ Probability that a from s leads to s’, i.e., P(s’| s, a)
◦ Also called the model or the dynamics
◦ A reward function R(s, a, s’)
◦ Sometimes just R(s) or R(s’) for simplicity
◦ A start state
◦ Maybe a terminal state
Double-Bandit MDP
Actions: Blue, Red
No discount
States: Win, Lose 0.25 $0 10 time steps
Both states have
the same value
0.75
$2
W 0.25 L
$0
$1 $1
0.75 $2
1.0 1.0
Online Planning
Rules changed! Red’s win chance is different.
?? $0

??
$2
W ?? L
$0
$1 $1
?? $2
? ??
Reinforcement Learning
Still assume a Markov decision process (MDP):
◦ A set of states s  S
◦ A set of actions (per state) A
◦ A model T(s,a,s’)
◦ A reward function R(s,a,s’)

Still looking for a policy (s)

New twist: don’t know T or R

◦ I.e. we don’t know which states are good or what the actions do
◦ Must actually try actions and states out to learn
New Game Agent
Reinforcement Learning
Deep
Mind -
AlpaGo
DeepMind Atari (© Two Minute
Lectures)

35
AlphaStar
 Robotics
Robotics
◦ Part mech. eng.
◦ Part AI
◦ Reality much harder than
simulations!
Technologies
◦ Vehicles
◦ Rescue
◦ Help in the home
◦ Lots of automation…
Images from UC Berkeley, Boston Dynamics, RoboCup, Google
Human-aid Interaction
Limitations
Narrowed models
Requires data and supervision
Can’t think like human
Safety, Ethics, and Privacy issues
Security
Cost and maintenance
Etc.
Utility
Clear utility function Not so clear utility function
Tools for Predictions & Decisions
Data
Definition
◦ Individual facts, statistic, or items of information

◦ In computing, data is information that has been translated into a form

that is efficient for movement or processing

42
Data explosion
Moore’s Law
Digital storage becomes larger, cheaper, and faster with each
successive year
Since 1986, the amount of data storage dramatically 8000

increases 7000

◦ 1986 2.6 EB 6000

◦ 1993 15.8 EB 5000

◦ 2000 54.5 EB 4000

◦ 2007 295 EB 3000

2000
◦ 2014 5000 EB
1000
◦ 2020 6800 EB
0
1980 1985 1990 1995 2000 2005 2010 2015 2020 2025

43
Data
Data Age
Data sources
◦ New York Stock exchange – 4-5 terabytes per day
◦ Facebook – more than 240 billion photos
◦ Cloud
◦ IoT …

IDC forecasts 175 zettabytes by 2025

◦ 90 zettabytes from IoT

Forbes predicts 150 trillion gigabytes of real-time data by 2025

44
Big data
RDBMS for big data
Big data
◦ Larger, more complex data set
◦ New data sources
◦ Semi or unstructured data

Big data is too large for traditional data processing software

45
Hadoop history
Started in 2002
◦ Doug Cutting and Mike Cafarella worked on Apache Nutch search engine project
funded by Yahoo
◦ They found that traditional database is too expensive for search engine to store
large volume of data

In 2003
◦ Google published GFS (Google File System) paper
Doug Cutting
In 2007
◦ Successfully tested Hadoop on 1000 node cluster

In 2008
◦ Yahoo released Hadoop as an open-source project to Apache Software
Foundation
◦ Successfully tested on 4000 node cluster
◦ Hadoop won Terabyte Sort Benchmark (209 seconds)

46
Reference: https://www.geeksforgeeks.org/hadoop-history-or-evolution/
Hadoop
Traditional approach
◦ Use local storage to store and access the data
◦ As the data size bigger, this approach is very complex and expensive

Hadoop approach
◦ In stead of local machines to fetch the data, it uses multiple nodes
◦ Each node holds part of the data

47
 Machine Learning
Up until now: use a model to make optimal decisions

Machine learning: how to acquire a model from data / experience

◦ Learning parameters (e.g., probabilities)
◦ Learning structure (e.g., BN graphs)
◦ Learning hidden concepts (e.g., clustering, neural nets)
What is Machine Learning?
Field of study that gives computers the ability to learn without
being explicitly programmed. – Arthur Samuel, 1959

A computer program is said to learn from experience E with

respect to some task T and some performance measure P, if its
performance on T, as measured by P, improves with
experience E. –Tom Mitchel, 1998
Machine Learning
Consider spam filter
Traditional approach
◦ Observe typical words or phrases (e.g., “4U”, “credit card”, “free”,
etc.), senders name, email body, or other patterns
◦ Write detection algorithm for each of patterns you found to identify
the spam
◦ Repeat above steps until it is good enough
Machine Learning
Consider spam filter
Machine learning approach
◦ Programmer design the model
◦ Machine learning model will process the data and find the
patterns shown in spams
◦ Machine learning keeps learning from data
Supervised vs. Unsupervised
Supervised learning Unsupervised learning
◦ Common technique for training ◦ Unlike supervised, this type of
AI system using large number of learning tries to identify patterns
labeled data set in data such as similarities to
◦ Uses huge amount of annotated categorize the data
data
◦ This learning seeks for grouping
◦ This type of system sometimes
requires fairly large amount of or clustering the data
data
◦ Problem is how to get a good
labeled data
AI, ML, and DL

Artificial Machine
Intelligence Learning Deep
Learning
Simulation of human Learning describes an
intelligence in machines to automatic search Extract
mimic human behavior process for better patterns
representations
 Deep Learning

Reference: https://towardsdatascience.com/why-deep-learning-is-needed-over-traditional-machine-learning-1b6a99177063
Why Deep Learning?
Manually feature extraction may easily cause over-specified, incomplete
and take a long time to design and validate
Can we find appropriate features from the data directly?
Deep learning provides a very flexible and universal
Works both unsupervised and supervised
Effective end-to-end joint system learning
Utilize large amounts of training data
Traditional ML vs. Deep Learning
Neural networks
Artificial network consists of artificial neurons or nodes
Each neurons or nodes are processing units
They are communicating each other with a weights
Each unit performs a relatively simple job
◦ Test: take input and compute an output propagating to next unit
◦ Learning: adjusting the weights
Neural Network
Inputs are feature values
Each feature has a weight
x1 w1
Sum and apply activation function
w2
Generate output x2 ෍ ∫ 𝑦ො

wn

…
xn

Inputs Weights Sum Activation Output

References
Book: AI, A Modern Approach by Russel and Norvig
AIMA of book: http://aima.cs.berkeley.edu/
AI in UC Berkeley: ai.berkeley.edu
Machine learning: Stanford
MIT Deep Learning note
Machine Learning Mastery: https://machinelearningmastery.com/
https://towardsdatascience.com/why-deep-learning-is-needed-over-traditional-machine-
learning-1b6a99177063
https://en.wikipedia.org/wiki/Activation_function
References
https://towardsdatascience.com/activation-functions-neural-networks-1cbd9f8d91d6
https://www.researchgate.net/figure/Description-of-deep-neural-network-DNN-model-a-Typical-structure-
of-DNN-model-b_fig3_339600446
Kourou, Konstantina, Themis P. Exarchos, Konstantinos P. Exarchos, Michalis V. Karamouzis, and Dimitrios I.
Fotiadis. "Machine learning applications in cancer prognosis and prediction." Computational and structural
biotechnology journal 13 (2015): 8-17.
https://towardsdatascience.com/gradient-descent-animation-1-simple-linear-regression-e49315b24672
https://mlfromscratch.com/optimizers-explained/#/

https://medium.com/ml-research-lab/under-fitting-over-fitting-and-its-solution-dc6191e34250