Machine Learning
for Everyone
Agenda
Goal: Encourage you to use Machine Learning… today!
‣ Aboutme
‣ MachineLearning
Misconceptions
Concepts
Problems andAlgorithms
‣ ForEveryone!
About me
Electronics Engineering, Software
Development, Data Science… Why not?
Neural TB
Tool that aids in the
diagnosis ofTuberculosis
using Neural Networks
Neural Ringer
Algorithm for online
electron/jet
discrimination forthe
ATLAS detector at CERN
using Neural Networks
djBrazil
Intelligent music platform
specialized in Brazilian
music
Jigsaw Dots
Interactive exploratory
visualization of
employees based on their
skills
Higgs Challenge
Machine
Learning
It is all about learning
Misconceptions
Too difficult
Big upfront investments
Needs supercomputers
Only for PhDs from MIT
Data Silo
Takes too long to pay off
ƃ
Reality
♥
Feature Extraction
Item {
Feature 1
Feature 2
Feature 3
…
Feature N
Feature2
0
8
16
24
32
40
Feature 1
0 10 20 30 40 50 60 70 80 90
Supervised Learning
23, 45, 67, 78
12, 48, 68, 22
…
34, 58, 77, 19
3
2
…
5
20, 39, 59, 68 3
♥
Items Features Labels Algorithm
New Item PredictionFeatures Model
Supervised Learning Algorithms
K-Nearest Neighbors Neural Networks
DecisionTree Random Forest
Regression
921 37
23 2487 1541
21 2121 21
?
Boston housing prices
Prediction of house
prices at Boston suburbs
based on census data
using Linear Regression
Classification
BA B
A AB AB
A AB B
?
Detecting particles
Online electron detection
based on more than 1500
detector cells using
Neural Networks
(GeV)TE
0 10 20 30 40 50 60 70 80
o(%)a
~
Prob.deRejeic
20
30
40
50
60
70
80
90
100
Ringer
T2Calo
smicosoC
η
-2 -1 0 1 2
o(%)a
~
Prob.deRejeic
80
85
90
95
100
Ringer
T2Calo
φ
-3 -2 -1 0 1 2 3
o(%)a
~
Prob.deRejeic
90
92
94
96
98
100
Ringer
T2Calo
Classifying products
Product classification on
9 different classes based
on 90 numerical features
usingAmazon Machine
Learning
Diagnosing Tuberculosis
Tuberculosis diagnosis
based on patients
questionnaires using
Neural Networks
Unsupervised Learning
23, 45, 67, 78
12, 48, 68, 22
…
34, 58, 77, 19
20, 39, 59, 68
♥
Items Features Algorithm
New Item
Better
RepresentationFeatures Model
Unsupervised Learning Algorithms
K-Means Self-Organising Maps t-SNE
Clustering
Clustering crime
Crime clusters based on
information from the San
Francisco Police
Department Crime
Incident Reporting
system using K-means
Dimensionality Reduction
Visualizing employees
Visualization of 2000+
employees described by
200+ skills after reducing
dimensionality using the
t-SNE algorithm
For everyone!
You can do it!
Massive Online Open Courses
Open Source Tools
Pay-as-you-go
Amazon
Machine Learning
Google Cloud
Machine Learning
Azure
Machine Learning
Cheat sheets
Kaggle
Sponsors
Workflow Reasons
Code Snippet
>>> classifier = RandomForestClassifier().fit(features, labels)
>>> prediction = classifier.predict(new_features)
23, 45, 67, 78
12, 48, 68, 22
…
34, 58, 77, 19
A
B
…
B
20, 39, 59, 68 A
♥
Items Features Labels Algorithm
New Item PredictionFeatures Model
Thank you.
Twitter: @dhianadeva
Email: dhiana@spotify.com

Machine Learning for Everyone