Artificial Intelligence

Lecture 9
IntelligentAgents, &
AI Related Disciplines
Artificial
Intelligence
Related
Disciplines

3
MainResourcesforthisSection

The following courses:

o Data Science, Harvard University
(School of Engineering & Applied
Sciences)
o Introduction to Data Science - Data
Intensive Computing, University of
Florida
o Introduction to Data Science, University
of California - Berkeley
o Practical Data Science: Introduction,
Carnegie Mellon University

4
Learning Agents?
What is a learning agent?
• They are agents that have the capability to change themselves.
They change themselves to improve the performance and it is
measured using a performance function
• Each agent has a goal to reach and has a performance measure.
The agent updates and enhances himself to achieve this task.
(increase the performance measure or score)

How does the learning agent do that?

• By giving him learning or training examples as given to the
students Or by a negative or a positive feedback (based an action
in the environment if the feedback is a reward (positive feedback)
or a penalty (negative feedback)) as we took before in the
reinforcement learning.

- Wolfgang Ertel, "Introduction to Artificial Intelligence," 2nd Edition (2017) 5

Learning Agents?

"Of particular interest in AI are Learning agents, which are

capable of changing themselves given training examples or

through positive or negative feedback, such that the average

utility of their actions grows over time."

6
 Solution is Machine Learning (ML)?
• Any software even if it interface agent (has no physical appearance
or a robot has not physical appearance). It should have the ability to
learn using feedback (reinforcement learning) or using different
examples he learn from it.

• This is called machine learning and it is a subfield from the AI (give

the software or robot the ability to learn from different examples or
reinforcement learning).

• A recent appeared science called deep learning is part of the

machine learning and it is part of the AI.

• AI started at the beginning of 50’s, while the machine learning

started at the beginning of 80’s, whereas, the deep learning at 2010.

7
Artificial Intelligence Vs. Machine
Learning Vs. Deep Learning ..
A formal short answer:

Deep Learning is a part of Machine Learning, which is a part of AI.

8
Artificial Intelligence Vs. Machine
Learning Vs. Deep Learning ..

9
 Machine Learning Vs. Deep Learning ..
• There are great difference between machine learning and deep
learning. Deep Learning is based on several algorithms, but it is
commonly based on artificial neural networks (ANN).

• ANN is one of the approaches in Machine learning.

• The ANN or the Network that deep learning is based on is very

deep and huge in terms of nodes, layers, weights and
parameters and its numbers.

The shape of the NN in the deep learning different definitely from

the NN in the ANN.(
Traditional ANN: Typically consists of one input layer, one hidden layer, and one output layer.
It is shallow and suited for simpler tasks.
Deep Learning NN: Contains multiple hidden layers (deep architecture), enabling it to learn
more complex patterns and hierarchical features from data, often used in tasks like image
recognition and natural language processing.)
10
Machine Learning Vs. Deep Learning ..

11
 Machine Learning Vs. Deep Learning ..
What is feature extraction?
• If you have a group of objects and you want to divide them into two
classes (car or not a car)
• In Machine learning: we do a step called feature extraction (This step in
machine learning is not done manually(by ML Algorithm), but it is
engineered manually).

• Use ML algorithm looks for the most important features to extract from
the input (it can be image, signal, text) then ML algorithm put code,
equations to extract these features.
• The result is in the form of number of numerical data and these data are
input to the model. (Features can be wheels, windows, a specific color).
• For example (In the Face detection or Face recognition problem) you need
to extract features such as (color of the eye, distance between nose,
mouth, and so on). To differ between other different faces.

• Try to find an algorithm to extract the requested features and then get the
numeric values and input them to the classification step.
• Feature extraction and classification steps are performed separately and12 not
merged together.
 Machine Learning Vs. Deep Learning ..
• Deep Learning gets the data and input them to one big model that
performs both feature extraction and classification at the same
time.
• You do not know the features extracted and it is the job of the
deep learning model. Also, the step of training and decision which
is known as a classification step is performed in the same stage.

• In deep learning the feature extraction and classification steps are

merged together.

• Deep Learning performs automatic feature extraction in contrast

with ML

13
 Machine Learning
• You have training data (text, document, audio, signals, images,
videos).
• We extract features as the second step if based on ML.
• The features are input to another ML and then the ML trains on it.
• There is a label that determine the class of the data.

• Face → Class 0
• Not a Face → Class 1
• Cancer → Class 0
• Non a cancer → Class 1

• A new input is added to test the model (predicative model) to

generate the class. The input data is not seen before.
• This is supervised Learning. It can be (Binary class or multi-class)
• The quality of predictions depends on the accuracy of the feature
extraction, the amount of training data, and the effectiveness of the
14
model used.
Machine Learning?

Generic Machine Learning Models

14
 Machine Learning
(Supervised Learning paradigm)
ML can be divided into classical learning (supervised or unsupervised)
Supervised Learning: is a type of learning that is based on labeled
inputs.
Supervised → (Classification or Regression)
• Classification: is the process of finding or discovering a model or
function which helps in separating the data into multiple categorical
classes. (Classify binary or multi-class). To build a software to classify
one or two or more classes. Algorithms: (Naïve Bayes, SVM, KNN,
decision trees, and MLP(Multi-Layer Perceptron)).
➢ (Cat or dog) classification algorithm.
➢ I can predict that the weather is cloudy, sunny, cold.
➢ Student will succeed or fail.
• Regression: is the process of finding a model or function for
distinguishing the data into continuous real values. Predict a number
not a class. Algorithms: (linear, polynomial, Ridge/lasso) regression
➢ I can predict or compute a temperate value.
➢ Student will get a 90% percent for example. 16
 Machine Learning (Unsupervised and
Reinforcement Learning paradigms)
Unsupervised Learning is based on unlabeled inputs.
Unsupervised Learning → (Clustering or Dimensionality Reduction)
• Clustering: Collect the similar features of objects or data in one group.
Algorithms: (K-means, Fuzzy means, Mean-shift)

• Dimensionality Reduction: You have data or input data with so many

information (some of the information are interesting or relevant and
some are not) We need to extract and include the relevant only.
➢ Here are some algorithms that do that with the aim of extracting
the relevant and reduce the amount of data to prevent distortion.
Algorithms: (t-SNE, PCA, and, LDA).

Reinforcement learning → Learning based on the feedback to do an

action and wait for the feedback to do the next action
Algorithms: (Genetic algorithm, Q-learning)
17
 Machine Learning (Ensemble Methods)
• Ensemble Methods → Is to combine or ensemble different techniques
together to enhance the overall performance.
➢ You can make a system to differ between cancer and not cancer.
The system is based on SVM , decision trees, KNN. Each one has an
accuracy. Can I mix all the three together to enhance the
performance? Methods: Yes using stacking, bagging, and boosting.
➢ Algorithms on boosting: (XG-boost, CatBoost, Adaboost, Light
Gradient Boosting Machine (light GBM).
➢ Algorithms on bagging : Random Forest is a combination of a group
of random trees to obtain a robust accuracy.

18
 Neural Nets and Deep Learning
• One of the most important approaches in machine learning is Neural
Networks: Deep learning is based on it.
• It started with a small models such as perceptron (It is traditional
structure to neural networks) (Single layer neurons algorithms: Single
perceptron, Adaline(Adaptive Linear Neuron), Hebbian rules) or
sometimes (Multi-layer neurons: Multi-layer perceptron (MLP), self-
organization map(SOM)).
• Then, the neural networks has become bigger and the architecture of it
also changed and then we have the concept of deep learning.
• Various types of deep learning networks exists such as convolutional
neural networks (CNN), recurrent neural networks (RNN) examples on it
are (Long term short memory (LSTM) algorithms, gated recurrent unit
(GRU) algorithms, LSM), generative adversarial networks (GAN), and
autoencoders.
19
Machine
Learning?

{Artificial
Intelligence}
Machine
Learning Map

19
 Machine Learning Approaches & their
corresponding Applications
• There exist several approaches in ML. Each approach has different types
of algorithms. Each approach is used in specific problems or topics.
Supervised Learning
• Classification: Used in diagnosis (cancer or not) (grade of cancer
(0,1,2,3) Image classification (breast cancer or not , Alzheimer or not).
• Regression: Used in Weather forecasting, Market forecasting, life
expectancy (patient will live how many days or months based on his
health situation, predict population growth (in 5, 10, 20 years), and
stock market.
Unsupervised Learning
• Clustering: Used in recommendation system, amazon, OTLOB,
Souq.com, (Something to buy, books to read, Films to see).
• Dimensionally reduction: used in big data visualization ( you need to see
the most important information). To visualize the most relevant
features.
Reinforcement learning: is used more in games or AI-based games,
robotics, and so on. 5
Machine Learning
Approaches & their
corresponding
Applications
21
 What about Data-Mining?
Machine learning and data mining often employ the same
methods and overlap significantly. They can be roughly
distinguished as follows:
o Machine learning focuses on prediction, based on known
properties learned from the training data.
o Data mining focuses on the discovery of (previously) unknown
properties in the data. This is the analysis step of Knowledge
Discovery in Databases.
You have data and you do not know the features or the properties of these
data. Then, you min in the data to search for knowledge and relations that
didn’t appear before.
Example: In supermarket some one bought a product and another one
bought same product. You can use data mining techniques to discover why
both bought the product. What is the relation between them.

7
What about Data-Mining?

Much of the confusion between these two research communities

comes from the basic assumptions they work with: in machine
learning, performance is usually evaluated with respect to the
ability to reproduce known knowledge, while in Knowledge
Discovery and Data Mining (KDD) the key task is the discovery
of previously unknown knowledge.

Data Mining depends on (Unsupervised learning to perform

knowledge discovery).
Most approaches in Data Mining is in the unsupervised learning
category in terms of techniques and algorithms

8
Artificial Intelligence Vs. Data-Science

DataScience-Some PossibleDefinitions:
Data Science is the science which uses computer science,
statistics and machine learning, visualization and human-
computer interactions to collect, clean, integrate, analyse, visualize,
interact with data to create data products.

Data science = statistics + data processing + machine learning +

scientific inquiry + visualization + business analytics + big data + …

24
WhyDataScience?

in 20th Century Innovations ..

Engineering and Computer Science played key role:
o Cars
o Airplanes
o Power grid
o Television
o Air conditioning and central heating
o Nuclear power
o Digital computers
o The internet

25
WhyDataScience?

But how about these 20th and 21th Century questions?

o Does fertilizer increase crop yields?

o Does Streptomycin cure Tuberculosis?

o Does smoking cause lung-cancer?

26
WhyDataScience?
What is the difference? .. Data

o Does fertilizer increase crop yields?

Answer: Collect and analyse agricultural experimental data.

o Does Streptomycin cure Tuberculosis?

Answer: Collect and analyse randomized trials data.

o Does smoking cause lung-‐cancer?

Answer: Collect and analyse observational studies data.

✓ Deductive versus empirical ..

✓ Solutions deduced mostly from theory versus solutions deduced
from mostly from data .. 27
The Dawn of Big Data

To understand the
phenomenon that is big data,
it is often described using five
Vs: Volume, Velocity, Variety,
Veracity, and Value.

Recently, Visualization,
Virality, & Viscosity were
added (thus, Eight Vs) ..

13
WhyDataScience?
To Understand Big Data.

You have data generated with a large amount that was not found
before.
• Volume: Has big volume (before we had data in bits, byte, mega)
Now we have data in (tera and peta bytes)

• Velocity: generating the data is fast and very fast (generate

mega byte of data can take months, years). While generating
giga bytes of data can take 1 hour or a couple of hours

29
 WhyDataScience?
• Variety: The data were converted into tabular forms (you have
fields and records and you store them only).
• You have images generated in different forms and formats on
facebook, twitter, Instagram, with our camera, security camera.
• You have audio from different mikes and telephone lines. in
different formats
• You have text (sms, message, in whatsapp, facebooks, tweets, text
in electronic newspaper). All the books has become text with
different formats.
• Old data varies in only 10%.
• In fact, 80% of the world’s data is now unstructured.

Data now has become veracity (messy)

You cannot trust it and its quality, accuracy and you are not sure of
that
Can I get a value from these data?
This what the big data trying to handle it.
30
TheDawnofBigData
o Volume refers to the vast amounts of data generated every second.
o Velocity refers to the speed at which new data is generated and the speed
at which data moves around.
o Variety refers to the different types of data we can now use. In fact, 80%
of the world’s data is now unstructured, and therefore can’t easily be put
into tables (think of photos, video sequences or social media updates).
o Veracity refers to the messiness or trustworthiness of the data. With
many forms of big data, quality and accuracy are less controllable.
o Value; It is all well and good having access to big data but unless we can
turn it into value it is useless.

16
 WhyDataScience?
• An example not very recent a survey from about 2 years
Volume of the data is very large
• In North America more than 3500 peta byte of data
• In Latin America more than 50 peta byte of data
• In Africa more than 200 peta byte of data
• In Japan more than 400 peta byte of data

Velocity
• You have more than 2.9 million email sent every second
• You have more than 20 hours videos uploaded every min
• You have 50 million tweets per day.

32
 WhyDataScience?
Variety
• Most of the data goes in people to people (web blogs, social
networks)
• People deal with machine (data are stored on smart cards, medical
devices, archives, digital TV, bank cards)
• Machine to Machine (sensor report to another machine, barcode read
and store on computer, surveillance camera and store the capture to
PC or computer)
• (scanner scan and send the data to another machine)

What are the values that I can get or produce from data that have this
volume, Varity, and velocity

• You can find that 49% productivity increase and sales increase $9.6B
• Increase in the productivity in the air transportation by 21% and sales
increase by $4.3B.

33
 Success Stories & a few data science examples …

MoneyBall ?!

Starting around 2001, the Oakland A’s picked players that scouts thought were no good
but data said otherwise.
20
Success Stories & a few data science examples …

Poverty Mapping

Abelson, Varshney, and

Sun. “Targeting Direct
Cash Transfers to the
Extremely Poor,” 2012

21
 Exercises
What have we learned?
▪ Define the following terms: (1) Learning Agents, (2) Machine Learning.
▪ Compare between the following: Machine Learning & Deep Learning.
▪ Draw a block diagram that describes a generic Supervised Machine Learning model.
 Exercises
What have we learned?
▪ What are the characteristics of Big data (the 5 Vs)? Describe briefly.
▪ Compare between the following: (1) Machine Learning & Data-mining
▪ Which Machine Learning approach is employed when realizing each the following
applications:
▪ Estimating Life expectancy.
▪ Customer Segmentation.
▪ Robot Navigation.
▪ Rcommender Systems.
▪ Market or Weather Forecasting.
▪ Medical Image Classification.
THANK YOU