Demystifying AI, Machine Learning and Deep Learning

© 2017 MapR Technologies
Applying Machine Learning to IOT:
End to End Distributed Pipeline for Real-
Time Uber Data Using Apache APIs: Kafka,
Spark, HBase
Carol McDonald
@caroljmcdonald

Agenda
•  What is AI?
•  Why now?
•  What is Machine Learning?
–  Examples
•  What is Deep Learning?
–  Examples

What is AI?

AI NSA MIT Late 80s

Problems with hard coded Rules
•  Rules are manual, uses a human expert
–  difficult to maintain
–  give a one size fits all decision! (2 times overdose same as 38 times)
•  Machine learning uses data and statistics
–  can give sorted probabilty, can precisely match/target individuals

What is Machine Learning?
Data Build ModelTrain Algorithm
Finds patterns
New Data Use Model
(prediction function)
Predictions
Contains patterns Recognizes patterns
f(X)

Why all the buzz now?
What has changed?

What has changed in the past 10 years?
Distributed computing
Streaming analytics
Improved machine learning

Distribute Computation
Driver sends
Program tasks
Data Distributed
across Cluster
Result

Apache Spark Distributed Datasets
Distributed Dataset
Node
Executor
P4
Node
Executor
P1 P3
Node
Executor
P2
partitioned
Partition 1
8213034705, 95,
2.927373,
jake7870, 0……
Partition 2
8213034705,
115, 2.943484,
Davidbresler2,
1….
Partition 3
8213034705,
100, 2.951285,
gladimacowgirl,
58…
Partition 4
8213034705,
117, 2.998947,
daysrus, 95….
•  Data read into Memory Cache
•  Partitioned across a cluster
•  Operated on in parallel
•  Cached in memory for iterations

Streaming Analytics

GPUs speed up Multi core servers for parallel processing
Cluster of GPUs 1 million times faster than Cray-1

Mythbusters explain Parallel graphics with GPU vs Sequential CPU
•  Painting a smily face with a sequential paint gun

Mythbusters explain Parallel graphics with GPU
•  Painting a smiling face with one blast from a parallel paint gun !

Machine Learning

Types of Machine learning

Supervised Machine Learning
Supervised
•  Classification
–  Naïve Bayes
–  SVM
–  Random Decision
Forests
•  Regression
–  Linear
–  Logistic
Machine Learning
Unsupervised
•  Clustering
–  K-means
•  Dimensionality reduction
–  Principal Component
Analysis
–  SVD
Label

Supervised Algorithms use labeled data
Data
features
Build Model
New Data
features
Predict
Use Model
X1, X2
Y
f(X1, X2) =Y
X1, X2
Y

ML Discovery Model Building
Model
Training/
Building
Training
Set
Test Model
Predictions
Test
Set
Evaluate Results
Historical
Data
Deployed
Model
Insights
Data
Discovery,
Model
Creation
Production
Feature Extraction
Feature
Extraction
●  Churn Modelling
Uber
trips
Stream
TopicUber
trips
New Data

Supervised Machine Learning: Classification & Regression
Classification
Identifies
category for item

Classification: Definition
Form of ML that:
•  Identifies which category an item belongs to
•  Uses supervised learning algorithms
–  Data is labeled
Sentiment

If it Walks/Swims/Quacks Like a Duck …… Then It Must Be a Duck
swims
walks
quacks
Features:
walks
quacks
swims
Features:

Debit Card Fraud Example
•  What are we trying to predict?
–  This is the Label or Target outcome:
–  Fraud or Not Fraud
•  What are the “if questions” or properties we can use to predict?
–  These are the Features:
–  Is the amount spent today > historical average?
–  Unusual region for card history ?
–  Known merchant or not ?

Decision Tree For Classification
•  Tree of decisions about features
•  Estimates IF THEN ELSE questions
•  Gives probability of a correct decision
Is the amount spent in 24
hours > average
Is the number of
states used from > 2
Are there multiple
Purchases today from
risky merchants?
YES NO
NoYES
Fraud
90%
Not Fraud
50%
Fraud
90%
Not Fraud
30%
YES No

Real Time Credit Card Fraud Detection with Apache Spark Streaming
1.  Get event credit card
transaction data
2.  Read card holder profile
3.  Calculate history
features
4.  Publish Alerts for fraud
and enriched events
https://mapr.com/blog/real-time-credit-card-fraud-detection-apache-spark-and-event-streaming/

Classification Identifies Category
•  Classification:
–  identifies which category a new item belongs to
•  Who will ( buy, churn, get admitted to hospital ) ?
•  What is the mood of this comment?
•  Retail Example:
–  Which promotion draws more customers ?
•  Healthcare Example:
–  Suggest Patient diagnosis
–  Identify patients with high readmission risk

Label
Probabilty
of Fraud 1
X
Features: trans amount, type of store,
Time Location difference last trans.
Fraud
0
Not Fraud
.5
Classification Probability Logistic Regression Example
Predicts probability an item belongs to a category

Supervised Learning: Classification Probability
•  Logistic Regression (and other algorithms) :
–  Predicts probability an item belongs to a category (eg probability of fraud)
•  What is probablity someone will ( buy, churn, get admitted to hospital ) ?
•  Probability customer will renew service
•  Healthcare:
–  Probability of readmission

Label:
Price of house
Y
X1, X2
Features: square feet,
number bedrooms, location
Data point: sum of x,
price
Sales price = intercept + coeff * X1 + coeff2 * X2
Regression Predicts Amount, Estimates relationship between X & Y

Regression Predicts by estimating the relationship between variables
•  Regression predicts a numeric value (eg price)
•  What will be the ( revenue, product demand , sales , # churners)
•  Retail Example:
–  Sales based on an event
•  Healthcare Example:
–  Days of hospital stay

What is Unsupervised Machine Learning?
Machine Learning
Unsupervised
•  Clustering
–  K-means
•  Dimensionality reduction
–  Principal Component
Analysis
–  SVD
Supervised
•  Classification
–  Naïve Bayes
–  SVM
–  Random Decision
Forests
•  Regression
–  Linear
–  Logistic

Unsupervised Algorithms use Unlabeled data
Customer GroupsBuild ModelTrain Algorithm
Finds patterns
New Customer
Purchase Data
Use Model
Similar Customer Group
Contains patterns Recognizes patterns
Customer purchase
data

Unsupervised Machine Learning: Clustering
Clustering
group news articles into different categories

Unsupervised Learning
Learning structure from unlabeled examples
NBA Players
http://www.sloansportsconference.com/wp-content/uploads/2012/03/Alagappan-Muthu-EOSMarch2012PPT.pdf

Clustering: Definition
•  Groups objects into clusters of high similarity
–  Customer segmentation
–  Text categorization
–  recommendations
•  Anomaly detection: find what’s not similar
x
x
x
x
x

Clustering Groups objects into Clusters of high similarity
•  What are the groups of (customers, patients..) with
similar (bevahior, purchases, symptoms, illness…)
•  Healthcare:
–  Patient similarity
•  Retail:
–  Group customers by purchases.

Bank Customer Segmentation: Bank Products, Card Purchases

Association, Co-Occurrence, Market Basket Recommendations
•  Retail
–  Products which are purchased
together
•  Take action:
–  Store layouts
–  Which products to put on
specials, promote, coupons…
•  Healthcare
–  Patients like mine cohorts

Deep Learning

Deep Learning
Multilayered neural networks

The Network is trained with images

Neural network neuron or node
Each node takes input data and a weight and outputs a confidence score to the next
layer

Each node outputs a confidence score to the next layer

Errors are calculted at the output layer

Errors are sent back through the network

This process is repeated, adjusting weights, until correct

This process is repeated with lots of images

Deep Learning
During this process layers learn the optimal features for the model

Deep Learning Features
•  Advantage:
–  Features do not have to be
predetermined
•  Disadvantage:
–  Decisions are a black box
Feature
Decisions
?

Deep Learning in the News!
FINANCE AUTON. DRIVING HEALTHCARE VOICE RECOG.
3/27/17 - Hedge funds
have been trying to
teach computers to
think like traders for
years. (Bloomberg)
4/3/17 – Daimler… to
deploy autonomous
taxis that customers
can hail using a
smartphone app by
the start of the next
decade. (Fortune)
3/28/17 - deep learning
is being applied to
processing medical
images … eye disease
… skin cancer (MIT
tech review)
3/31/17 - IBM research
… advancing speech
recognition by applying
deep learning into
acoustic and lang.
models (InfoQ)

Deep Neural Networks
•  Classification and
•  Forecasting
Deep
Neural
Networks

Convolutional Neural Networks for Images
•  Insights from image & video files
Convolutional
Neural
Networks

Ex. PATIENT MORTALITY PREDICTION
1Scientific RepoRts | 7: 1648 | DOI:10.1038/s41598-017-01931-w
www.nature.com/scientificreports
Precision Radiology: Predicting
longevity using feature engineering
and deep learning methods in a
radiomics framework
LukeOakden-Rayner1,2
,GustavoCarneiro3
,Taryn Bessen1
, JacintoC. Nascimento4
,Andrew P.
Bradley5
& Lyle J. Palmer2
Precision medicine approaches rely on obtaining precise knowledge of the true state of health of an
individual patient, which results from a combination of their genetic risks and environmental exposures.
This approach is currently limited by the lack of effective and efficient non-invasive medical tests to
define the full range of phenotypic variation associated with individual health. Such knowledge is
critical for improved early intervention, for better treatment decisions, and for ameliorating the steadily
worsening epidemic of chronic disease.We present proof-of-concept experiments to demonstrate how
routinely acquired cross-sectionalCT imaging may be used to predict patient longevity as a proxy for
overall individual health and disease status using computer image analysis techniques. Despite the
limitations of a modest dataset and the use of off-the-shelf machine learning methods, our results are
comparable to previous ‘manual’ clinical methods for longevity prediction.This work demonstrates
that radiomics techniques can be used to extract biomarkers relevant to one of the most widely used
outcomes in epidemiological and clinical research – mortality, and that deep learning with convolutional
neural networks can be usefully applied to radiomics research.Computer image analysis applied
to routinely collected medical images offers substantial potential to enhance precision medicine
initiatives.
Measuring phenotypic variation in precision medicine
Precision medicine has become a key focus of modern bioscience and medicine, and involves “prevention and
treatment strategies that take individual variability into account”, through the use of “large-scale biologic data-
bases … powerful methods for characterizing patients … and computational tools for analysing large sets of
data”1
. The variation within individuals that enables the identification of patient subgroups for precision medicine
strategies is termed the “phenotype”. The observable phenotype reflects both genomic variation and the accumu-
lated lifestyle and environmental exposures that impact biological function - the exposome2
.
Precision medicine relies upon the availability of useful biomarkers, defined as “a characteristic that is objec-
tively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or phar-
macological responses to a therapeutic intervention”3
. A ‘good’ biomarker has the following characteristics: it is
sensitive, specific, predictive, robust, bridges clinical and preclinical health states, and is non-invasive4
.
Genomics can produce good biomarkers useful for precision medicine5
. There has been significant success in
exploring human genetic variation in the field of genomics, where data-driven methods have highlighted the role
of human genetic variation in disease diagnosis, prognosis, and treatment response6
. However, for the chronic
and age-related diseases which account for the majority of morbidity and mortality in developed nations7
and
worldwide8
, the majority (70–90%) of observable phenotypic variation is related to non-genetic determinants9
.
1
Department of a io o o a e ai e ospita ort errace e ai e 5000 ustra ia. c oo of u ic
ea t e ni ersit of e ai e ort errace e ai e 5000 ustra ia. 3
c oo of omputer cience e
ni ersit of e ai e ort errace e ai e 5000 ustra ia. 4
Instituto uperior cnico is on ortu a .
5
c oo of Information ec no o an ectrica n ineerin e ni ersit of ueens an ui in 78 t ucia
D 40 7 ueens an ustra ia. orrespon ence an re uests for materia s s ou e a resse to .O. emai :
u eoa enra ner mai .com)
Received: 8 December 2016
Accepted: 6 April 2017
Published: xx xx xxxx
OPEN
Oakden-Rayner, et al.,
Scientific Reports, May 2017
com/scientificreports/
Figure 4. Images at the level of the proximal left anterior descending coronary artery, with the most strongly
predicted mortality and survival cases selected by averaging the predictions from the deep learning and
engineered feature models. The mortality cases (left side) demonstrate prominent visual changes of emphysema,
cardiomegaly, vascular disease and osteopaenia. The survival cases (right side) appear visually less diseased and
frail.
Mortality Survival

Example: Exploiting Unstructured Data
http://www.economist.com/news/science-and-technology/21664943-computers-can-
recognise-complication-diabetes-can-lead-blindness-now - Sep 19, 2015
Diabetic Retinopathy:
•  Challenging to diagnose from
image (84% consensus)
•  Crowd-sourced to Kaggle
•  Deep-learning and convolutional
NN used to classify image data
•  Winning model showed 85%
accuracy rate

Recurrent Neural Networks for Sequenced data
•  Sequence of events and language
applications
Recurrent
Neural
Networks

To Learn More:
•  MapR Quick Start solutions
https://mapr.com/solutions/big-data-and-hadoop-quick-start-solutions/
•  Customer 360, Recommendation Engine, Log Analysis, Risk, Deep Learning

MapR Deep Learning QSS
New Image
to Classify
Category
Probabilities
Training
Images…
Category
1
Category
N
…
MapR-FS
MapR Data Platform
Kubernetes
Enterprise Storage Database Event Streaming
MapR-FS MapR-DB MapR Streams
Global Namespace High Availability Data Protection Multi-tenancy Unified Security
D
MapR Converged Data Platform
POD 1
DD MASTER
NODE
POD 2 POD 3
Parameter
Server 1
TF Trainer
1
TF Trainer
2

Fit your business model
Common Use Cases
•  Churn prediction
•  Customer clustering
•  Product recommendation
•  Budget optimization
•  ETA
•  Sales prediction
•  Pricing model
•  …
Cost function -- real business impact
•  Leverage A/B testing

90+% of effort is logistics,
not learning

Big Data – Machine Learning Cycle
Big
Data
Identify a problem
Prepare Data Model Data Get Insight
Test a Solution
EvaluateMonitor Deploy
Machine LearningReference: head of Machine learning at Uber

End to End Streaming Analytics Example Application
https://mapr.com/blog/monitoring-real-time-uber-data-using-spark-machine-learning-streaming-and-kafka-api-part-1/

MapR Blog
• https://www.mapr.com/blog/

…helping you put data technology to work
●  Find answers
●  Ask technical questions
●  Join on-demand training course
discussions
●  Follow release announcements
●  Share and vote on product ideas
●  Find Meetup and event listings
Connect with fellow Apache
Hadoop and Spark professionals
community.mapr.com

We reinvented the data platform
for next-gen intelligent applications & Data Science
On-Premise, In the Cloud, Hybrid
NoSQL Webscale
Storage
MessagingMultiple
Processing
Engines
Real Time Unified Security Multi-tenancy Disaster Recovery
Streaming
Multiple compute engines and tools operating concurrently
Immediate access to vast amounts of diverse data
Low latency for millisecond responsiveness
Support diverse workloads simultaneously
Able to be a reliable system of record
Enterprise grade reliability

Q&A
ENGAGE WITH US

Demystifying AI, Machine Learning and Deep Learning

More Related Content

What's hot (19)

Similar to Demystifying AI, Machine Learning and Deep Learning (20)

More from Carol McDonald (13)

Recently uploaded (20)

Demystifying AI, Machine Learning and Deep Learning