INTRODUCTION TO ML

Learning 2

 Learning is the process of acquiring new understanding, knowledge, behaviors,

skills, values, attitudes, and preferences — Wikipedia
 The new understanding of the rules governing the processes of universe allows us
to extrapolate and make decisions in the unforeseen situations
 The ability to learn is possessed by humans, animals, and some plants.
 Learning by humans has a very vast scope as it covers everything in the universe
o Recognize objects
o Compute
o Design

 Machine learning (ML) intends to make machines to learn in a very limited scope

Machine Learning
What Does it Mean to Learn? 3

 Alice has just begun taking a course on machine learning. What do we expect when we need her to “learn”
all about this topic.
 What are the steps involved:
o Teaching through examples

o Exam to gauge learning

 She has done well at learning if she does well on the exam. But what makes a reasonable exam?
o If Bob spends the entire semester talking about machine learning, and then gives Alice an exam on

History of Pottery, then Alice’s performance on this exam will not be representative of her learning.
o On the other hand, if the exam only asks questions that Bob has answered exactly during lectures,

then this is also a bad test of Alice’s learning, especially if it’s an “open notes” exam.
o What is desired is that Alice observes specific examples from the course, and then has to answer

new, but related questions on the exam.

 Generalization is perhaps the most central concept in learning and machine learning.
Machine Learning
A Concrete Example 4

 Consider a course recommendation system for undergraduate computer science students.

 Each student has taken, and evaluated, a subset of the courses. The evaluation is simply a score from -2 (terrible) to +2
(awesome).
 The job of the recommender system is to predict how much a particular student (say, Alice) will like a particular course
(say, Algorithms).
 Given historical data from course ratings (i.e., the past) we are trying to predict unseen ratings (i.e., the future).
o Asking whether Alice is likely to enjoy the History of Pottery course
is unfair because the system has no idea
what History of Pottery even is, and has no prior experience with this course.
 we’re expecting the system to generalize beyond its experience, which is unfair
o On the other hand, we could ask it how much Alice will like Artificial Intelligence, which she took last year and
rated as +2 (awesome).
o We would expect the system to predict that she would really like it, but this isn’t demonstrating that the system

has learned: it’s simply recalling its past experience

 we’re not expecting it to generalize at all.
 This general set up of predicting the future based on the past is at the core of most machine learning.
Machine Learning
Basic Machine Learning Idea 5

 Receive a collection of observations associated with some action label

 Perform some kind of “Machine Learning”
o to be able to:

 Receive a new observation

 “Process” it and generate an action label that is based on previous observations
 Main Requirement: Good generalization
 Teaching a child to recognize an apple
o Showing an apple to a child

o Telling her that is an apple

Machine Learning
What is ML? 6

 Layman Definition
o The science (and art) of programming computers so they can learn from data.

 General Definition
o The field of study that gives computers the ability to learn without being explicitly

programmed.
—Arthur Samuel, 1959
 A more engineering-oriented definition:
o 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 Mitchell, 1997
Machine Learning
What is ML? 7

Regardless of the definition you choose, at its most basic level, the goal
of machine learning is to adapt to new data independently and make
decisions and recommendations based on thousands of calculations and
analyses. 

Machine Learning
Computer Program 8

 Analogous to a mathematical function

o Takes input

o Applies transformation

o Gives output

 In terms of computer program

o Input Data
calcPrice(item, count, discount){
o Transformation Set of rules unitPrice = readPrice(item)
itemPrice = unitPrice*(1-discount)
netPrice = count * itemPrice
o Output Result }
return netPrice

Machine Learning
ML vs Traditional Programming 9

In traditional programming, Answers

Rules
focus is on getting answers from Machine Learning
Data
rules and data
Rules In machine learning, focus is
Traditional Answers
on establishing the rules by
Data Programming providing a lot of data and
corresponding answers

Why do we need to flip the programming paradigm?

Machine Learning
 Representative ML Problems 10

Digit Recognition: It is very hard to say what makes a 2

If(speed ==0)
Status = Rest
else if(speed > 0 && speed <= 4)
Status = Walking
else
Status = running
How to specify the rules to recognize
cycling or golfing etc?

Machine Learning
The Flipped Programming Paradigm of ML? 11

 It is very hard to write programs that solve problems like recognizing a face.
o We don’t know what program to write because we don’t know how our brain does it.
o Even if we had a good idea about how to do it, the program might be horrendously

complicated.
 Instead of writing a program by hand, we collect lots of examples that specify the correct
output for a given input.
 A machine learning algorithm then takes these examples and produces a program that does
the job.
o The program produced by the learning algorithm may look very different from a

typical hand-written program. It may contain millions of numbers.

o If we do it right, the program works for new cases as well as the ones we trained it on.

Machine Learning
Not All Problems are Suited for ML 12

Sort these numbers in

decreasing order

2, 4, 18, 1, 77, 0, 85

Does not make sense.

Makes sense.
Do not use Machine
Use Machine Learning
Learning

(what are the risks if you

do?)

Machine Learning
Why ML Shines? 13

The traditional approach The Machine Learning approach

Automatically adapting to change

Machine Learning
Candidate Problems for ML 14

 Human expertise do not exist

o Navigating on Mars

 Applications in which the data and hence the rules keep changing
o Spam Filter

 Spammers keep changing the techniques

 Will need to keep updating the rules till eternity
 Problems that either are too complex for traditional approaches or have no known algorithm.
o Speech Recognition

 Thousands of words spoken by millions of people

 Will need millions of rules to even recognize a small subset of words
 Problems that involve discovery of new knowledge from large databases (data mining).
o Market basket analysis (e.g. diapers and beer)

o Medical text mining (e.g. migraines to calcium channel blockers to magnesium)

Machine Learning
Why is ML Necessary? 15

 Learning is a hallmark of intelligence;

o many would argue that a system that cannot learn is not intelligent.

 Without learning, everything is new;

o a system that cannot learn is not efficient because it rederives each

solution and repeatedly makes the same mistakes.

 Why is learning possible?
o Because there are regularities in the world.

Machine Learning
Resurgence of Interest in ML 16

 Nearly limitless quantity of available data,

 Affordable data storage,
 The growth of less expensive and more powerful processing
 More robust machine learning models capable of analyzing bigger and more
complex data while delivering faster, more accurate results on vast scales
 Availability of ML tools enable organizations to more quickly identify
profitable opportunities and potential risks.

Machine Learning
SUCCESS STORIES OF ML
Insurance Industry 18

 Major Use Cases

o Auto, Property &Life…  Data
 Pricing o Customer
 Risk Management  Demographics and historical record
 Claim Loss Prediction o Property
 Claim Loss Prevention  Actual property data, local/regional data
 Customer Experience o Streaming
 Fraud Detection
 Near real time personal state and behavioral
data ranging from social sentiment analysis
to biometrics, and related news and weather

Machine Learning
AXA Insurance: Auto Insurance Pricing 19

 Technical Approach
o Using TensorFlow on the Google Cloud

Machine Learning Engine

 Input Layer
o Age,

o Address,

o Vehicle type,

o Previous Accidents,

o Original Channel,

o Car Age Range

 Predicted Risk Accurately ~78%

o Cost optimization & new services

Machine Learning
Pharma 20

 Major Use Cases

o Drug discovery

o Clinical Trials

o Analysis

(Biochemical)
 Data
o Biochemistry

o Clinical Trial

Case Data
Drug Discovery & Clinical Trials
o Journals, News…

Machine Learning
Pharma 21

Optimizing Virtual Screening With Machine Learning

Machine Learning
Pharma 22

IBM PATENTS DRUG DISCOVERY MACHINE LEARNING MODELS

Machine Learning
Healthcare 23

 Major Use Cases

o Epidemic Prediction & Monitoring
o Diagnosis

o Treatment

 Data
o Patient

 Demographics and historical record/EHR

o Treatment/Outcome Data

 Taxonomies, drug interaction guides, journals, case records…

 Streaming
o Near real time personal state and behavioral data ranging from social sentiment analysis to

biometrics, and related news and weather

Machine Learning
Healthcare 24

INFECTIOUS DISEASE MANAGEMENT

Machine Learning
Healthcare 25

CANCER PREDICTION AND PROGNOSIS

Machine Learning
Healthcare 26

 Retinal damage caused by diabetes, which may lead to

blindness
 Accounts for 12% of new cases of blindness in the US
annually
 Leading cause of blindness for ages 20-64
 At least 90% of new cases could be reduced with proper
treatment and monitoring
 Affects up to 80%of people with diabetes for 20+ years
 Often has no early warning signs
 Non-proliferative diabetic retinopathy (NPDR) is the first
stage
 Detected by fundus photography to see micro aneurisms DIABETIC RETINOPATHY DIAGNOSIS
Machine Learning
Healthcare 27

 “Our technology combines two various

analytics approaches into one hybrid method,
wherein a convolutional neural networks
(CNN)- based method for DR classification is
integrated with a dictionary based learning
that incorporates DR-specific pathologies.
 This hybrid analysis resulted in a great
improvement in classification accuracy.
 Our method takes approximately twenty
seconds to analyze the image and achieves an
accuracy score of 86 percent in classifying
the disease across the five severity levels.

DIABETIC RETINOPATHY DIAGNOSIS

Machine Learning