Machine Learning

Problems, Financial
Domain Applications &
Introduction to
Optimization
By Rahul Rai,
Dean’s Distinguished Professor
Clemson University
Founder and CEO, AI Brilliance

www.aibrilliance.com
 HELLO,

I am Dr. Rahul Rai,

Dean’s Distinguished Professor
at Clemson University
Greenville, South Carolina, United States

Founder and CEO

Data Science & Machine Learning | Artificial Intelligence | Diagnostics &

Prognostics | Computer Vision & Autonomous Systems | Digital Manufacturing
& Design | Geometric & Shape Reasoning | AR/VR/XR | Applied Ontology
 Connect/Follow @ LinkedIn
Search: Rahul Rai Clemson
University

I direct the Geometric Reasoning

and Artificial Intelligence Lab (GRAIL)

I earned my doctoral degree in Mechanical Engineering

from The University of Texas at Austin USA in 2006
focusing on Sequential Sampling techniques.
Sequential sampling is popularly known as “Active
Learning” in the machine learning domain.
Poll
What is your level of expertise in machine
learning?

Choose from Below

A. None

B. Beginner

C. Intermediate

D. Expert
We Will Cover

What is machine
Learning (ML) ?

Types of machine
learning problems

Introduction to
optimization concept
What Is Deep Learning

A technique that enables

machines to mimic human
behavior

Subset of AI which uses

statistical methods to enable
machines to improve at tasks
with experience

Subset of ML based on neural

networks that permits a
machine to train itself to
perform a task
Machine learning is programming
computers to optimize a performance
criterion using example data or past
experience.

-Ethem Alpaydin
The goal of machine learning is to develop methods
that can automatically detect patterns in data, and
then to use the uncovered patterns to predict
future data or other outcomes of interest.

-Kevin P. Murphy
The field of pattern recognition is concerned with
the automatic discovery of regularities in data
through the use of computer algorithms and with
the use of these regularities to take actions.

- Christopher M. Bishop
Machine Learning:
Multiple Names
§ Data mining
ML applied to “databases”, i.e.
collections of data

§ Inference and/or estimation in statistics

§ Pattern recognition in engineering

§ Signal processing in electrical

engineering

§ Induction
Machine Learning:
Adopted Concepts

Data
Mining
statistics Control
Theory

information Decision
theory Theory
Machine
Learning

databases Cognitive
Science

evolutionary Neuro-science
models
Machine Learning Is
Step 01

LEARN
Training
Model/
Data Predictor

Past

Step 02

PREDICT
Testing
Data Model/
Predictor

Future
 Main Steps of Machine Learning Applications
AutoML (computer program)

Model Deployment
Problem Data Feature Optimization Evaluation
Definition

Data Collection Feature Engineering Model Selection Algorithm Selection

Applying to Real Applications

Yao et al., 2018, Taking the Human out of Learning Applications: A Survey on Automated Machine Learning
Classes of Machine
Learning Approaches

Machine Learning

Supervised Task Driven

Unsupervised Data Driven

Reinforcement Learning From

Mistakes
 Supervised Learning:
Classification
Classification is a class of supervised learning:
Given a set of finite labels ,
Used to predict group membership for data
instances

Apple Apple Banana Banana

LABEL

MODEL PREDICTOR
Supervised Learning:
Classification

MODEL PREDICTOR

PERDICTED
LABEL

Supervised learning: learn to

predict new example
 Supervised Learning:
Classification
Machine learning techniques used to rank a set of
inputs according to relevance

1 2 3 4

Query Label

Ranking: Label is ranking

Supervised Learning:
Ranking
Given a query and a set of web pages, rank them
according to relevance.

01. Google Search

Supervised Learning:
Ranking
Given a query a watched movie, rank other movies in
the databased according to relevance.

02. Netflix Recommendations

Supervised Learning:
Regression
Machine learning techniques used to find a well
defined and known function underlying your
data.

125 160 80 65

MODEL PREDICTOR

Regression: label is real-valued

Supervised Learning:
Regression

MODEL PREDICTOR

150
predicted maximum weight

Supervised learning: learn to

predict new example
Unsupervised Learning

Unsupervised learning: given data,

i.e. examples, but no labels
Unsupervised Learning:
Clustering
Clustering: model the underlining structure or
distribution in the data without labels

Apple Apple Banana Banana

MODEL PREDICTOR
Unsupervised Learning:
Clustering

MODEL PREDICTOR

Unsupervised learning: learn to

cluster new example
Better Performance for Free

• unlabeled data is cheap

• labeled data can be hard to get
• human annotation is boring
• labels may require experts
Semi-Supervised Learning

Using both labeled and unlabeled data

to build better learners, than using
each one alone.

Banana

MODEL PREDICTOR
Machine Learning
Applications in Finance
Algorithmic Trading
§ Prediction of stock or index prices
§ Prediction of forex or cryptocurrency trends

Risk Assessment
§ Bankruptcy prediction
§ Credit scoring
§ Credit evaluation
§ Loan/Insurance underwriting

Fraud Detection
§ Text-based information retrieval
Machine Learning
Applications in Finance
Portfolio Management
§ Portfolio optimization
§ Portfolio selection
§ Portfolio allocation

Financial Sentiment Analysis

§ Sentiment extraction through social media feeds

Financial Text Mining

§ Credit Card Fraud
§ Money laundering
§ Tax evasion
§ Bank fraud
Event-Driven Stock Trend
Prediction
Input
§ Long-Term events 𝑈 !
§ Mid-Term events 𝑈 "
§ Short-Term events 𝑈 #

Output
§ Stock trend (increase or decrease)

Increase or Decrease Output Layer

Hidden Layer

𝑈! 𝑈" 𝑈# Input Layer

Event-Driven Stock Trend
Prediction
How to process the long-term and mid-
term events?
output has same dimension with
short time events
Max pooling

Share same weights

long-term or mid-term events

Switch
Introduction to
optimization concepts

www.aibrilliance.com
Poll

What is your optimization concept

level?
Choose from Below

A. None

B. Beginner

C. Intermediate

D. Expert
Optimization Problem

Example:
Design of a thin wall tray with minimal material

VIDEO

https://www.youtube.com/watch?v=Df748qOXMvc
Optimization Problem

Example:
The tray has a specific volume, 𝑽, and a given
height, 𝑯.
The Design problem is to select the length , 𝒍,
and Width, 𝒘, of the tray.

𝒘
𝒍
Optimization Problem
Optimization Problem

Example:
The design is to minimize material volume, (or
weight), where “T” is an acceptable small value
for wall thickness.

𝒘
𝒍
Optimization Problem

objective function
Optimization Problem
Optimization Problem

𝒘
𝒍
Steepest Descent Method

1. Derivative information required

2. Derivatives must be available

3. Derivatives must be continuous

4. Gradients are analytically available

or reliable approximations are
numerically available
Steepest Descent Method
Optimization Problem

Example:

VIDEO

https://www.youtube.com/watch?v=gU7JCPSSMYc
Example of 2D Gradient
Example of 2D Gradient
Example of 2D Gradient
Example of 2D Gradient
Example of 2D Gradient
Steepest Descent Method
Steepest Descent Method

Downhill
Direction
Step Size

Step 1: Start with an initial guess

Step 2:
1. Determine descent direction
2. Choose a step size
3. Update
Step 3: Repeat step 2
Step Size Determination
Step size determination is also known as line search.

It is single variable problem ….you can use exact method…Conjugate

direction method…or a good algorithm for single variable
optimization….

Such as:
Steepest Descent Method

Basic Equation:
Steepest Descent Method

Example
Steepest Descent Method
Quiz Time
When the gradient descent method is
started from a point near the solution, it
will converge very quickly.

Choose from Below

A. True

B. False
Quiz Time
When the gradient descent method is
started from a point near the solution, it
will converge very quickly.

Choose from Below

A. True

B. False
Thank You
Connect/Follow @ LinkedIn
Search: Rahul Rai Clemson
University

www.aibrilliance.com