2. Course Information
¨ Instructor: Amir EL-Ghamry
¤amir_nabil@mans.edu.eg
¤amirelghamry@email.arizona.edu
¤amirnabilsaleh@gmail.com
¤Facebook @amir.n.saleh
¤Twitter @AmirNabilSaleh
¤Instagram @amirnabilsaleh
(C) Dhruv Batra
2
3. Course Information
¨ About Me:
¤Bachelor of computer science at
Mansoura university @2006
¤Grade : Excellent
¤Rank : First
¤Master degree @ 2012
¤Ph.D. degree @ 2019
(C) Dhruv Batra
3
4. Course Information
¨ About Me:
¤Demonstrator at CS dept 2007 – 2012
¤Assistant Lecturer at CS dept 2012 – 2019
¤Joint supervision mission to USA 2016 – 2018
¤University of Texas at Dallas
4
5. Course Information
¨ About Me:
¤Member of Open Event Data Alliance
(OEDA) Research project at the university of
Texas at Dallas 2017 – 2018
¤ Consultant of Minerva Research project at
Arizona University 2019 – till now
5
6. Syllabus
¨ Basics of Statistical Learning
nLoss functions, bias-variance tradeoff,
overfitting, cross-validation
¨ Supervised Learning
nNearest Neighbour, Naïve Bayes, Logistic
Regression, Support Vector Machines,
Neural Networks, Decision Trees
6
8. Syllabus
¨ You will learn about the methods you
heard about
¨ You will understand algorithms, theory,
applications, and implementations
¨ It’s going to be FUN and HARD WORK J
8
9. Prerequisites
¨ Probability and Statistics
¨ Calculus and Linear Algebra
¨ Algorithms
¨ Programming
¤Python or Your language of choice for project.
¨ Ability to deal with abstract mathematical
concepts
9
10. Textbook
¨ We will have lecture notes.
¨ Reference Books:
¤ [Online]
Machine Learning: A Probabilistic Perspective
Kevin Murphy
¤ [Free PDF from author’s webpage]
Bayesian reasoning and machine learning
David Barber
http://web4.cs.ucl.ac.uk/staff/D.Barber/pmwiki/pmwik
i.php?n=Brml.HomePage
¤ Fundamental of neural network by laurene fausett
10
25. Learning Data
25
• Training set
– Used to fit (train) model parameters
• Validation set
– Used to check performance on independent
data and tune parameters
• Test set
– final evaluation of performance after all
parameters fixed
27. Supervised Learning
27
A Technique that build the classifier Automatically
• Create classifier by finding patterns in image
Collect
training data
Train
Classifier
Make
Predictions
33. Changing the Training data
33
Classifying the car based on Horsepower and
number of seats
34. Machine Learning is better
34
• You can create new classifier for new problem
by just changing the training data instead of
Writing rules for each problem
• Reusable
35. Many types of Classifiers
35
• Artificial Neural Network (ANN)
• Support vector machine (SVM)
• Decision tree
• Random forest
• Naïve Bayes
• … etc
36. Good features
36
Classifier is good as the feature you provide
è So
Coming up with good features is one of the
most important jobs in Machine Learning
39. Good features – sample population
39
• Population of 1000 dogs
• Number of greyhounds = 500.
• Number of Labradors = 500.
• We Draw a histogram for their heights
• greyhounds color : red
• Labradors color : blue
41. Good features – sample population
41
If we want to predict the dog with heights
20 , 25 and 35 inches
• Dogs with 20 inches è Labrador
• Dogs with 35 inches è greyhound
42. Good features – sample population
43
• Dog with 25 inches è ????
Probability that it is a greyhound
Or Labrador is very close
That mean:
Height is a useful feature but not
perfect
43. Good features
44
Conclusion :
In Machine Learning we need multiple
features for better results
If you want to know which features to use
Do a thought experiment