Machine Learning: an Introduction and cases

Disclaimer
Presentations are intended for educational
purposes only and do not replace independent
professional judgment. Statements of fact and
o p i n i o n s e x p r e s s e d a r e t h o s e o f t h e
participants individually and don’t necessarily
reflect those of blibli.com.
Blibli.com does not endorse or approve, and
assumes no responsibility for, the content,
accuracy or completeness of the information
presented.

Machine Learning:
an Introduction and cases
Hendri Karisma
hendri.karisma@gdn-commerce.com

Hendri Karisma
• Sr. Research and Development Engineer
at blibli.com (PT. Global Digital Niaga)
• R&D Team in AI Squad
• Working for Fraud Detection System,
Customer Group for abuser detection,
dynamic recommendation system project
and Customer Segmentation.
• https://about.me/hendriKarisma

Materials
• Definition and background
• Methods
• Problems and Solutions
• Technologies
• Cases

The Definition of Informatics
“Automation of Information” –
Prof. Dr. Ing. Iping Supriana

Artificial Intelligence
 S. Rusel and P. Norvig, Artificial
Intelligence in Modern Approach

Problem Solving agent
 Searching for solution
 Knowledge Base and Planning
 Reasoning
 Learning

Learning
Generalization
Specific Cases
Inductive LearningDeductive Learning

Machine Learning Definition
“A computer program is said to learn from
experience E with respect to some class of
tasks T and performance measure P, if its
performance at tasks in T, as measured by P,
improves with experience E.” – Prof. Tom Mitchel

Problem and Solutions
 Analytical (Exact)
Example :
— analytics solution :
— Numerical solution
— Error = | 7.25 – 22/3| = |7.25-7.33|=0.08333
 Numerical (Aprox)
— Is numerical methods just about ML method that we know in the book?
— Newton raphson, Gauss Elimination, Gauss-Jordan, Jacobi method, Gauss-
Seidel, Lagrange, Newton Gregory, Richardson Interpolation, etc.

How it works
Count the error ( y - y')
Then minimize the error
or
Maximize the likelihood

Machine Learning Function
 Information Theory (Decission Tree : ID-Tree,
C4.5, etc)
 Probability (Bayessian : Naive Bayes, Belief
Network, etc)
 Graphical Model (Belief network, HMM, CRF,
Neural Network, etc)
 Numerical Method / Regression (Stochastic
Gradient Descent: Linear Regression, Multiple
Linear Regression, Neural Network,
Stochastic Gradient Ascent : E-M Algorithm)

Machiner Learning Taxonomy
 Supervised
 Unsupervised
 Reinforcement Learning
 Semi-Supervised
 Deep Learning

Solutions #2
 In-memory data fabric: provides low-latency access and
processing of large quantities of data by distributing data
across the dynamic random access memory (DRAM), Flash, or SSD
of a distributed computer system

Solutions #3
 Cluster machine
 GPU Machine (OpenCL and nVidia CUDA)

Tools on python
 Numpy
 Scipy
 Pandas
 Scikit-learn
 Matplotlib
 Seaborn
 Tensorflow
 *pydata.org
 *anaconda
 Other Tech (to support
ML) :
– Apache Kafka
– Apache Spark
– Db: mongo,postgre
– elasticsearch
– CUDA/OpenCL

Tools on java virtual machine (jvm)
 Weka
 Deeplearin4j (working with spark and gpu)
 H2O (working with spark and GPU, support
tensor, mxnet, and cafe)
 JcuDNN (JNI wrapping nvidia cuDNN)
 Mahout (hadoop)
 Mllib Spark

How we applied Machine Learning

 We are using micro services
Another problem

What data that we need???
How to Collect

Event Drive Architecture
Service ServiceService Service Service
Message queue (kafka)
RecommedationFraud Service
Machine Learning Engine

Anomaly Detection
 Anomalies are patterns in data that do not conform to a well
defined notion of normal behavior
 These nonconforming patterns : outliers, discordant
observations, exceptions, aberrations, surprises, peculiarities, or
contaminants in different application domains

Challanges
 Defining a normal region
 In many domains normal behavior keeps evolving
 Availability of labeled data for training/validation of models used
by anomaly detection techniques is usually a major issue
 Often the data contains noise that tends to be similar to the
actual anomalies

Anomaly Detection
Solution Method :
 Gaussian MIxture Model
 Fitted by EM - Algorithm

Gaussian Distribution
Before GMM, try to remember the gaussian distribution

Gaussian Distribution Multivariate

Gaussian Mixture
We have 3 gausians
mean 3 clusters on the
right picture.
We have 4 gausians
mean 4 clusters on the
left picture.

Gaussian Mixture (Multivariate)

Gaussian Mixture Multivariate
We have more than one gaussian mixture mean we have more
than one possible position for each data that we want to
distribute to GMM.
for example we have data x then want to trying to distribute x
to GMM, then we need to calculate the probability of x in first
gaussian, then in second gaussian, until our last gaussian. It
mean we have p(x) given each gaussians parameters.
f(x) = argmax{p(x | μ1, Σ1), p(x | μ2, Σ2), p(x | μ3, Σ3),... p(x | μn+1, Σn+1) }
where p is :
p(x | μ, Σ)

EM (Expectation Maximization) Algorithm
• GMM Ref : https://brilliant.org/wiki/gaussian-mixture-model/

EM (Expectation Maximization) Algorithm
, Dimana

Fraud Detection
Payment Fraud (phishing,
account take-over, carding)
System abuse (promo, content,
account, logistic and payment
methods especially COD)
Fraud not only result in financial
losses but also produce some
reputational risk.
Some security measures has
been taken by bank or another
multinational finance service.
[E. Duman et al, 2013]

Machine Learning: an Introduction and cases

More Related Content

What's hot

Similar to Machine Learning: an Introduction and cases

More from Hendri Karisma

Recently uploaded

Machine Learning: an Introduction and cases