How Machines can take decisions
Viju Chacko
About Me
Viju Chacko
https://www.linkedin.com/in/vijuchacko/
viju.Chacko@gmail.com
IoT Practice Head (Utilities ISU) @ Tata Consultancy Services
Agenda
• Introduction
• Machine Learning Vocabulary
• Machine Learning approaches
• Other ML Techniques
• Introduction to Neural Networks
• Different types of Neural Networks
• GPGPU : General-Purpose Computation on GPU
• Other AI Components: Within an Enterprise
• Q&A
A forgotten definition of Computer
“Computer is an electronic device that process information as per predefined
instruction/rules”
A forgotten definition of Computer
“Computer is an electronic device that process information as per predefined
instruction/rules”
and with Artificial Intelligence
“Computers can also look at past instances of an event, look at the
inputs and outputs, and then predict the outputs for given inputs
without pre-defined instructions/rules”
Why Artificial Intelligence
is so relevant now?
• Large scale Storage and
Faster Compute (but
smaller in size) are possible, all
available over Cloud*
* Digital Evolution Revolution
Machine Learning in our daily lives
An approach to Machine Learning - Human
Intelligence
• How do we make decisions while driving the car?
• How does an experienced stock trader places his order?
• How do a doctor diagnosis your illness?
Explicitly or Implicitly human beings learn to read (and understand
the impact) of the parameters
Machine Learning Vocabulary
Sepal Length Sepal Width Petal Length Petal Width Species
6.7 3.0 5.2 2.3 Virginica
6.4 2.8 5.6 2.1 Virginica
4.6 3.4 1.4 0.3 Setosa
6.9 3.1 4.9 1.5 Versicolor
4.4 2.9 1.4 0.2 Setosa
4.8 3.0 1.4 0.1 Setosa
5.9 3.0 5.1 1.8 Virginica
5.4 3.9 1.3 0.4 Setosa
4.9 3.0 1.4 0.2 Setosa
5.4 3.4 1.7 0.2 Setosa
Features
Target
Label
Types of Machine Learning
Machine
Learning
Supervised:
Supervised learning is
the machine learning
task of inferring a
function from labeled
training data.
Regression: Outcome
is continuous
(numerical)
Classification:
Outcome is a
category
Unsupervised:
Unsupervised learning is
a type of
machine learning
algorithm used to draw
inferences from datasets
consisting of input data
without labeled
responses.
Ex: Categorizing the emails to Primary, Social,
Promotions, Updates, Forums
Ex: Predicting Stock price
Ex: Explore the set a given data set and identify
possible classifications within the data
Training and Test Data from history
Training
Data
Test Data
A very Basic Linear Regression
History
Input
(x)
Output
(y)
1 6
4 18
3 14
2 10
6 26
2 10
3 14
5 22
6 26
2 10
1 6
A very Basic Linear Regression
History
Input
(x)
Output
(y)
1 6
4 18
3 14
2 10
6 26
2 10
3 14
5 22
6 26
2 10
1 6
Current Input
Input
(x)
Output
(y)
9 ?
8 ?
1 6
9  38
8 34
A very Basic Logistic Regression
Input
(x)
Output
(y)
90 4.54E-05
91 0.000123
92 0.000335
93 0.000911
94 0.002473
95 0.006693
96 0.017986
97 0.047426
98 0.119203
99 0.268941
100 0.5
101 0.731059
102 0.880797
103 0.952574
104 0.982014
105 0.993307
106 0.997527
107 0.999089
108 0.999665
109 0.999877
110 0.999955
A very Basic Logistic Regression
Input
(x)
Output
(y)
90 4.54E-05
91 0.000123
92 0.000335
93 0.000911
94 0.002473
95 0.006693
96 0.017986
97 0.047426
98 0.119203
99 0.268941
100 0.5
101 0.731059
102 0.880797
103 0.952574
104 0.982014
105 0.993307
106 0.997527
107 0.999089
108 0.999665
109 0.999877
110 0.999955
Current Input
Input
(x)
Output
(y)
75 ?
210 ?
75  Out (1)
210  In (0)
Fitting the Model
Error measurement – Efficiency of a model
Life is not always simple equations
• Linear Regression
• Logistic Regression
• Polynomial Regression
• Stepwise Regression
• Ridge Regression
• Lasso Regression
• ElasticNet Regression
Applying mathematical/statistical approaches for finding the relationship
between inputs and outputs (Historical events)
Not just all .. there are other techniques as well
Support Vector Machine / Support Vector Classifier
Kernel Approximation : An approach of applying a simplification function over any input (A simple Definition)
Principal component analysis: An approach to take only the relevant features, or identify the relevant features
….
Ensemble Modeling- Just one model vs.
a bunch of them
It’s the system that
matters
If the bee disappeared off the surface of the globe,
then man would have only four years of life left. No
more bees, no more pollination, no more plants, no
more animals, no more man.
-- Albert Einstein
A closer look at another Machine
learning approach inspired from
this MACRO/MICRO universe
Neurons – Artificial / Natural
Activation
Function
A more closer look at an Artificial Neuron
x = inputs
w = weights
b = bias term
f(z) = output
Why Network of
Neurons?
• Real world problems cannot be mapped to
mathematical/statistical models
• Following the steps of nature
Feed forward neural network
Weights
Input Layer
Output Layer
Hidden layers
Training a Neural Network
Different types of Neural Networks
• Convolutional Neural Networks :A Convolutional neural network
(CNN, or ConvNet) is a class of deep, feed-forward artificial neural
networks that has successfully been applied to analyzing visual
imagery.
• Recurrent Neural Network(RNN): Hidden layers receive their own
outputs as input
• Long / short term memory (LSTM): An improvement over RNN
• Restricted Boltzmann machines (RBM),Deep Belief Networks (DBN) ,
Deep convolutional inverse graphics networks (DCIGN)….
*http://www.asimovinstitute.org/neural-network-zoo/
by Fjodor Van Veen
GPGPU : General-Purpose Computation on GPU
* Images are copyright material of subsequent brands
Other AI Components: Within an Enterprise
&
Questions?

How machines can take decisions

  • 1.
    How Machines cantake decisions Viju Chacko
  • 2.
    About Me Viju Chacko https://www.linkedin.com/in/vijuchacko/ viju.Chacko@gmail.com IoTPractice Head (Utilities ISU) @ Tata Consultancy Services
  • 3.
    Agenda • Introduction • MachineLearning Vocabulary • Machine Learning approaches • Other ML Techniques • Introduction to Neural Networks • Different types of Neural Networks • GPGPU : General-Purpose Computation on GPU • Other AI Components: Within an Enterprise • Q&A
  • 4.
    A forgotten definitionof Computer “Computer is an electronic device that process information as per predefined instruction/rules”
  • 5.
    A forgotten definitionof Computer “Computer is an electronic device that process information as per predefined instruction/rules” and with Artificial Intelligence “Computers can also look at past instances of an event, look at the inputs and outputs, and then predict the outputs for given inputs without pre-defined instructions/rules”
  • 6.
    Why Artificial Intelligence isso relevant now? • Large scale Storage and Faster Compute (but smaller in size) are possible, all available over Cloud* * Digital Evolution Revolution
  • 7.
    Machine Learning inour daily lives
  • 8.
    An approach toMachine Learning - Human Intelligence • How do we make decisions while driving the car? • How does an experienced stock trader places his order? • How do a doctor diagnosis your illness? Explicitly or Implicitly human beings learn to read (and understand the impact) of the parameters
  • 9.
    Machine Learning Vocabulary SepalLength Sepal Width Petal Length Petal Width Species 6.7 3.0 5.2 2.3 Virginica 6.4 2.8 5.6 2.1 Virginica 4.6 3.4 1.4 0.3 Setosa 6.9 3.1 4.9 1.5 Versicolor 4.4 2.9 1.4 0.2 Setosa 4.8 3.0 1.4 0.1 Setosa 5.9 3.0 5.1 1.8 Virginica 5.4 3.9 1.3 0.4 Setosa 4.9 3.0 1.4 0.2 Setosa 5.4 3.4 1.7 0.2 Setosa Features Target Label
  • 10.
    Types of MachineLearning Machine Learning Supervised: Supervised learning is the machine learning task of inferring a function from labeled training data. Regression: Outcome is continuous (numerical) Classification: Outcome is a category Unsupervised: Unsupervised learning is a type of machine learning algorithm used to draw inferences from datasets consisting of input data without labeled responses. Ex: Categorizing the emails to Primary, Social, Promotions, Updates, Forums Ex: Predicting Stock price Ex: Explore the set a given data set and identify possible classifications within the data
  • 11.
    Training and TestData from history Training Data Test Data
  • 12.
    A very BasicLinear Regression History Input (x) Output (y) 1 6 4 18 3 14 2 10 6 26 2 10 3 14 5 22 6 26 2 10 1 6
  • 13.
    A very BasicLinear Regression History Input (x) Output (y) 1 6 4 18 3 14 2 10 6 26 2 10 3 14 5 22 6 26 2 10 1 6 Current Input Input (x) Output (y) 9 ? 8 ? 1 6 9  38 8 34
  • 14.
    A very BasicLogistic Regression Input (x) Output (y) 90 4.54E-05 91 0.000123 92 0.000335 93 0.000911 94 0.002473 95 0.006693 96 0.017986 97 0.047426 98 0.119203 99 0.268941 100 0.5 101 0.731059 102 0.880797 103 0.952574 104 0.982014 105 0.993307 106 0.997527 107 0.999089 108 0.999665 109 0.999877 110 0.999955
  • 15.
    A very BasicLogistic Regression Input (x) Output (y) 90 4.54E-05 91 0.000123 92 0.000335 93 0.000911 94 0.002473 95 0.006693 96 0.017986 97 0.047426 98 0.119203 99 0.268941 100 0.5 101 0.731059 102 0.880797 103 0.952574 104 0.982014 105 0.993307 106 0.997527 107 0.999089 108 0.999665 109 0.999877 110 0.999955 Current Input Input (x) Output (y) 75 ? 210 ? 75  Out (1) 210  In (0)
  • 16.
  • 17.
    Error measurement –Efficiency of a model
  • 18.
    Life is notalways simple equations • Linear Regression • Logistic Regression • Polynomial Regression • Stepwise Regression • Ridge Regression • Lasso Regression • ElasticNet Regression Applying mathematical/statistical approaches for finding the relationship between inputs and outputs (Historical events)
  • 19.
    Not just all.. there are other techniques as well Support Vector Machine / Support Vector Classifier Kernel Approximation : An approach of applying a simplification function over any input (A simple Definition) Principal component analysis: An approach to take only the relevant features, or identify the relevant features ….
  • 20.
    Ensemble Modeling- Justone model vs. a bunch of them
  • 21.
    It’s the systemthat matters If the bee disappeared off the surface of the globe, then man would have only four years of life left. No more bees, no more pollination, no more plants, no more animals, no more man. -- Albert Einstein A closer look at another Machine learning approach inspired from this MACRO/MICRO universe
  • 22.
    Neurons – Artificial/ Natural Activation Function
  • 23.
    A more closerlook at an Artificial Neuron x = inputs w = weights b = bias term f(z) = output
  • 24.
    Why Network of Neurons? •Real world problems cannot be mapped to mathematical/statistical models • Following the steps of nature
  • 25.
  • 26.
  • 27.
  • 28.
  • 29.
  • 30.
  • 31.
    Different types ofNeural Networks • Convolutional Neural Networks :A Convolutional neural network (CNN, or ConvNet) is a class of deep, feed-forward artificial neural networks that has successfully been applied to analyzing visual imagery. • Recurrent Neural Network(RNN): Hidden layers receive their own outputs as input • Long / short term memory (LSTM): An improvement over RNN • Restricted Boltzmann machines (RBM),Deep Belief Networks (DBN) , Deep convolutional inverse graphics networks (DCIGN)…. *http://www.asimovinstitute.org/neural-network-zoo/ by Fjodor Van Veen
  • 32.
    GPGPU : General-PurposeComputation on GPU * Images are copyright material of subsequent brands
  • 33.
    Other AI Components:Within an Enterprise
  • 34.