Machine Learning
Cicek Guven, Itir Onal
This course covers four ML algorithms: Decision Tree, Perceptron, Logistic Regression and Neural
Networks. The first chapter is about Machine Learning in general. The fourth chapter covers the
optimization problem (decoupled from a model), which is applicable to any of the algorithms. The fifth
chapter discusses the optimal representation of data in a model.
LECTURE 1 INTRODUCTION ....................................................................................................................... 3
1.1 MACHINE LEARNING .......................................................................................................................................... 3
1.2 TYPES OF LEARNING PROBLEMS ............................................................................................................................ 4
1.3 EVALUATION: HOW WELL IS THE ALGORITHM LEARNING? .......................................................................................... 5
LECTURE 2 DECISION TREE......................................................................................................................... 7
2.1 LEARNING RULES WHILE PLAYING A GAME............................................................................................................... 7
2.2 HOW TO BUILD A DECISION TREE ........................................................................................................................... 8
2.3 EFFICIENCY, SPEED AND DEPTH OF A DT ............................................................................................................... 14
2.4 IMPURITY MEASURES ........................................................................................................................................ 16
2.5 HOW CAN WE USE DECISION TREES FOR REGRESSION?............................................................................................ 17
2.6 ADVANTAGES AND DISADVANTAGES OF DECISION TREES .......................................................................................... 17
LECTURE 3 PERCEPTRON ......................................................................................................................... 18
3.1 WHAT IS A PERCEPTRON?.................................................................................................................................. 18
3.2 ALGORITHM.................................................................................................................................................... 19
3.3 POSSIBLE STUMBLING BLOCKS ............................................................................................................................ 25
LECTURE 4 GRADIENT DESCENT ............................................................................................................... 30
4.1 INTRODUCTION ............................................................................................................................................... 30
4.2 ERROR AND OPTIMIZATION ................................................................................................................................ 31
4.3 SLOPE, DERIVATIVE AND GRADIENT...................................................................................................................... 33
4.4 STOCHASTIC GRADIENT DESCENT (SGD)............................................................................................................... 36
LECTURE 5 REPRESENTATION .................................................................................................................. 38
5.1 OVERVIEW OF ML PIPELINE ............................................................................................................................... 38
5.2 FEATURE ENGINEERING ..................................................................................................................................... 38
5.3 DOMAIN DEPENDENCE ...................................................................................................................................... 39
5.4 CHOOSING FEATURES ....................................................................................................................................... 40
5.5 TEXT CLASSIFICATION........................................................................................................................................ 46
5.6 IMAGE CLASSIFICATION ..................................................................................................................................... 49
5.7 FEATURE ABLATION ANALYSIS ............................................................................................................................. 52
,LECTURE 6 LOGISTIC REGRESSION............................................................................................................ 53
6.1 RECAP OF UPDATE RULE OF (S)GD ...................................................................................................................... 53
6.2 LOSS FUNCTION FOR CLASSIFICATION ................................................................................................................... 55
6.3 COST FUNCTION OF LOGISTIC REGRESSION ............................................................................................................ 63
6.4 SGD FOR THE LOSS FUNCTION ............................................................................................................................ 64
6.5 SUMMARY LINEAR REGRESSION VS. LOGISTIC REGRESSION ....................................................................................... 66
6.6 HOW TO CONTROL (OVER)FITTING? .................................................................................................................... 67
LECTURE 7 NEURAL NETS......................................................................................................................... 71
7.1 RECAP ........................................................................................................................................................... 71
7.2 THE BRAIN ...................................................................................................................................................... 73
7.3 FEED-FORWARD NEURAL NETWORK (A.K.A. MULTI-LAYER PERCEPTRON).................................................................. 74
7.4 NEURAL NETWORK PREDICTION.......................................................................................................................... 77
7.5 EX: REPRESENTING XOR .................................................................................................................................. 78
7.6 COST FUNCTIONS ............................................................................................................................................. 80
7.7 TRAINING THE NEURAL NETWORK ...................................................................................................................... 82
7.8 SUMMARY OF (ARTIFICIAL) NEURAL NETWORKS ..................................................................................................... 85
7.9 SPECIAL TYPES OF NEURAL NETWORKS ................................................................................................................ 86
,Lecture 1 Introduction
1.1 Machine Learning
Machine Learning (ML) is the study of computer algorithms that improve automatically through
experience. It involves becoming better at a task (T), based on some experience (E) with respect to some
performance measure (P).
1.1.1 Learning process
1) Find examples of labels/experiences.
2) Come up with a learning algorithm, which infers rules from examples (training set).
3) Applied the rules to new data.
1.1.2 Examples
- Filter email: If (A or B or C) and not D, then “spam”.
- Recognize handwritten numbers and letters.
- Recognize faces in photos.
- Determine whether text expresses positive, negative or no opinion.
- Guess a person’s age based on a sample of writing.
- Flag suspicious credit-card transactions.
- Recommend books and movies to users based on their own and others’ purchase history.
- Recognize and label mentions of people’s or organization names in text.
ML is not meant for random guessing, like predicting the number when rolling some dice. It studies
algorithms that learn from examples.
, 1.2 Types of learning problems
Type Input Response Example
Regression A (real) number predict person’s age, predict price of a stock,
predict student’s score on exam
Binary classification YES/NO answer (condition being there detect SPAM, predict polarity of product review:
or not there) positive vs negative
Multiclass One of a finite set of options detect species based on photo, classify newspaper
classification article as <politics> <sports> …
Multilabel A finite set of YES/NO answers assign songs to one or more genres (rock – pop –
classification metal, hip-hop – rap)
Ranking Object ordered according to relevance rank web pages in response to user query, predict
student’s preference for courses in a program
Sequence labeling a sequence of elements a corresponding sequence of labels label words in a sentence with their syntactic
(ex. words) category (noun – adverb – verb)
Sequence-to- a sequence of elements sequence of other elements (possibly translations (“My name is Penelope” → “Me
sequence modeling (ex. words) different length, possibly elements from llamo Penélope”), computer-generated subtitles
different sets)
Autonomous measurements from instructions for actuators (steering, self-driving car
behavior sensors (microphone, accelerator, brake …)
accelerometer …)
Cicek Guven, Itir Onal
This course covers four ML algorithms: Decision Tree, Perceptron, Logistic Regression and Neural
Networks. The first chapter is about Machine Learning in general. The fourth chapter covers the
optimization problem (decoupled from a model), which is applicable to any of the algorithms. The fifth
chapter discusses the optimal representation of data in a model.
LECTURE 1 INTRODUCTION ....................................................................................................................... 3
1.1 MACHINE LEARNING .......................................................................................................................................... 3
1.2 TYPES OF LEARNING PROBLEMS ............................................................................................................................ 4
1.3 EVALUATION: HOW WELL IS THE ALGORITHM LEARNING? .......................................................................................... 5
LECTURE 2 DECISION TREE......................................................................................................................... 7
2.1 LEARNING RULES WHILE PLAYING A GAME............................................................................................................... 7
2.2 HOW TO BUILD A DECISION TREE ........................................................................................................................... 8
2.3 EFFICIENCY, SPEED AND DEPTH OF A DT ............................................................................................................... 14
2.4 IMPURITY MEASURES ........................................................................................................................................ 16
2.5 HOW CAN WE USE DECISION TREES FOR REGRESSION?............................................................................................ 17
2.6 ADVANTAGES AND DISADVANTAGES OF DECISION TREES .......................................................................................... 17
LECTURE 3 PERCEPTRON ......................................................................................................................... 18
3.1 WHAT IS A PERCEPTRON?.................................................................................................................................. 18
3.2 ALGORITHM.................................................................................................................................................... 19
3.3 POSSIBLE STUMBLING BLOCKS ............................................................................................................................ 25
LECTURE 4 GRADIENT DESCENT ............................................................................................................... 30
4.1 INTRODUCTION ............................................................................................................................................... 30
4.2 ERROR AND OPTIMIZATION ................................................................................................................................ 31
4.3 SLOPE, DERIVATIVE AND GRADIENT...................................................................................................................... 33
4.4 STOCHASTIC GRADIENT DESCENT (SGD)............................................................................................................... 36
LECTURE 5 REPRESENTATION .................................................................................................................. 38
5.1 OVERVIEW OF ML PIPELINE ............................................................................................................................... 38
5.2 FEATURE ENGINEERING ..................................................................................................................................... 38
5.3 DOMAIN DEPENDENCE ...................................................................................................................................... 39
5.4 CHOOSING FEATURES ....................................................................................................................................... 40
5.5 TEXT CLASSIFICATION........................................................................................................................................ 46
5.6 IMAGE CLASSIFICATION ..................................................................................................................................... 49
5.7 FEATURE ABLATION ANALYSIS ............................................................................................................................. 52
,LECTURE 6 LOGISTIC REGRESSION............................................................................................................ 53
6.1 RECAP OF UPDATE RULE OF (S)GD ...................................................................................................................... 53
6.2 LOSS FUNCTION FOR CLASSIFICATION ................................................................................................................... 55
6.3 COST FUNCTION OF LOGISTIC REGRESSION ............................................................................................................ 63
6.4 SGD FOR THE LOSS FUNCTION ............................................................................................................................ 64
6.5 SUMMARY LINEAR REGRESSION VS. LOGISTIC REGRESSION ....................................................................................... 66
6.6 HOW TO CONTROL (OVER)FITTING? .................................................................................................................... 67
LECTURE 7 NEURAL NETS......................................................................................................................... 71
7.1 RECAP ........................................................................................................................................................... 71
7.2 THE BRAIN ...................................................................................................................................................... 73
7.3 FEED-FORWARD NEURAL NETWORK (A.K.A. MULTI-LAYER PERCEPTRON).................................................................. 74
7.4 NEURAL NETWORK PREDICTION.......................................................................................................................... 77
7.5 EX: REPRESENTING XOR .................................................................................................................................. 78
7.6 COST FUNCTIONS ............................................................................................................................................. 80
7.7 TRAINING THE NEURAL NETWORK ...................................................................................................................... 82
7.8 SUMMARY OF (ARTIFICIAL) NEURAL NETWORKS ..................................................................................................... 85
7.9 SPECIAL TYPES OF NEURAL NETWORKS ................................................................................................................ 86
,Lecture 1 Introduction
1.1 Machine Learning
Machine Learning (ML) is the study of computer algorithms that improve automatically through
experience. It involves becoming better at a task (T), based on some experience (E) with respect to some
performance measure (P).
1.1.1 Learning process
1) Find examples of labels/experiences.
2) Come up with a learning algorithm, which infers rules from examples (training set).
3) Applied the rules to new data.
1.1.2 Examples
- Filter email: If (A or B or C) and not D, then “spam”.
- Recognize handwritten numbers and letters.
- Recognize faces in photos.
- Determine whether text expresses positive, negative or no opinion.
- Guess a person’s age based on a sample of writing.
- Flag suspicious credit-card transactions.
- Recommend books and movies to users based on their own and others’ purchase history.
- Recognize and label mentions of people’s or organization names in text.
ML is not meant for random guessing, like predicting the number when rolling some dice. It studies
algorithms that learn from examples.
, 1.2 Types of learning problems
Type Input Response Example
Regression A (real) number predict person’s age, predict price of a stock,
predict student’s score on exam
Binary classification YES/NO answer (condition being there detect SPAM, predict polarity of product review:
or not there) positive vs negative
Multiclass One of a finite set of options detect species based on photo, classify newspaper
classification article as <politics> <sports> …
Multilabel A finite set of YES/NO answers assign songs to one or more genres (rock – pop –
classification metal, hip-hop – rap)
Ranking Object ordered according to relevance rank web pages in response to user query, predict
student’s preference for courses in a program
Sequence labeling a sequence of elements a corresponding sequence of labels label words in a sentence with their syntactic
(ex. words) category (noun – adverb – verb)
Sequence-to- a sequence of elements sequence of other elements (possibly translations (“My name is Penelope” → “Me
sequence modeling (ex. words) different length, possibly elements from llamo Penélope”), computer-generated subtitles
different sets)
Autonomous measurements from instructions for actuators (steering, self-driving car
behavior sensors (microphone, accelerator, brake …)
accelerometer …)
