AI Made Easy For All
Fundamentals of AI for Non-ICT Professionals
Motivation
Some Important
Trends
Establishment of a
Dedicated Facility for AI
Machine Learning For Professionals: Foundation
Supervised Learning Unsupervised Learning Reinforcement Learning
Prof. Dr. S. Mohsin Ali Kazmi
Director Continuing Engineering Education and
Training Center
Introduction
• Learning
• Daily Life Example
• Machine Learning Definition
• Machine Learning Time Lines
• Why Machine Learning
• Machine Learning Components
• Data
• Statistical Analysis
• Model Development
• Model Validation
“Learning is any process by
which a system improves
performance from experience”
Herbert A. Simon
American Economist, Cognitive Psychologist
Among Pioneers of AI
Turing Award: 1975
Nobel Prize: 1978
Daily Life Learning
Interactive session
- How we Buy and Who we Consult:
- Car
- House
- Furniture
- Experienced persons are living examples of machine learning algorithms
which develop expertise from training sets consisting of life experiences
to give valuable advices
Machine learning
Computer Programs that
automatically improve their
performance through experience
But this is the second thing.
☺
Interactive session
Machine Learning
Deep Learning
Why Machine Learning
• Develop systems that can automatically adapt and customize
themselves to individual users
• Discover New Knowledge from large database
• Automate monotonous tasks which requires some intelligence
Data in context
• Flood of Available Data
• Increasing Computational Power
• Industrial Competitive Demands
• Rapid Development in Algorithms
Components of Machine
Learning
Machine Learning Mechanism
Statistical Supervised
Training Data Learning
Analysis
Machine
Large
Model
Development Learning
Pool
Of
Unsupervised
Learning
Basics
Data
Model
Testing Data
Validation
Machine
Learning
Reinforcement
Classifications
Prediction Learning
AB+
Made of Words Shoe
• Gender Size
• Eye Color Fraudulent
Ordinal Data has hierarchy Customer
Qualitative Data
100
Nominal Data has no hierarchy Ton/hr No. of
Large Kids
Pool
Of
Data
Pain
Continuous Infinite Options Severity
Quantitative Data
pH = 9
Discrete Finite Options
Made of Numbers
Customer
• Age
Satisfaction
• No. of Kids
70% / 30% OR 80% / 20% OR 90% /10%
70% OR 80% OR 90%
Statistical
Training Data
Analysis
Large
Pool
Of
Data
Testing Data
Range
Statistical
Training Data
Analysis
Large
Pool Model
Of Development
Data Y= mx + C
Linear Algorithms Y = aX1+bX2+C
A Mathematical
relationship between
independent and
Testing Data
dependent variables. Non-Linear
Algorithms aX2+bX+c = 0
Statistical
Training Data
Analysis
Large
Pool Model
Of Development
Data
Testing Data Model Validation 30% OR 20% OR 10%
Error Estimation= Prediction - Actual
Training Data Statistical Analysis
Large
Pool Model
Of Development
Data
Testing Data Model Validation
Prediction
Machine Learning Mechanism Machine Learning Classifications
Both Training and
Statistical Supervised Testing Data are
Training Data Learning
Analysis available
Large
Pool Model
Of Development
Data
Unsupervised
Learning
No Training Data
Model
Testing Data
Validation
Reinforcement
Prediction Learning Environment Interactive Learning
• Forecasting
Regression • Process Optimization
Statistical Supervised
Training Data Learning
Analysis
Classification • Fraud/Fault Detection
• Customer Retention
Large
Pool Model
Of Developmen
Data t
Unsupervise
d Learning
Model
Testing Data
Validation
Reinforcement
Prediction Learning
ENGINEERING AND MACHINE LEARNING
Grey:Ideal Case for Machine Learning
Blue: Chemical Engineering Research offers Transformation
Red: Not useful for Machine Learning; Need engineering
methods to generate more data.
APPLICATIONS
• Operational Optimization
• Composition Estimation
• Fault Detection (Heat Exchanger Cleaning, Filter switchover, Process Drift)
• Catalyst development
• General Design of Experiment
• Modelling for Process Control (Predictive)
Thank You
Introduction to
Generative AI
(GenAI) & Large
Language
Models (LLMs)
Mr. Kamran Ali
The Power of
Generative AI:
Creating New
From Existing
Generative AI is a field of This content can be text, Generative AI models
AI focused on creating images, video, audio, learn patterns from
new content code, and more existing data
They then use these The building block of Gen
patterns to generate AI is what we call Large
entirely new content Language Models (LLMs)
Demystifying LLM:
What is it
&
How it learns to speak
our Language
A large language model is a type
of Artificial Intelligence algorithm
that uses deep learning techniques
and massively large data sets to
understand, summarize, generate
and predict new content
LLM Development Timeline
1967 1997 2010 2011 2017 2020
• ELIZA • LSTM • Stanford • Google • Google • OpenAI
CoreNLP Brain Transformer GPT-3
2021 2022 2023 2024
• Google • OpenAI • OpenAI • OpenAI
LaMDA GPT-3.5 GPT-4 GPT-4o
• Google • Meta • Google
PaLM LLAMA Gemini
LLM Learning Method
LLMs use a Deep learning involves
technique called artificial neural
deep learning networks inspired by
the human brain
Networks consist With training, LLMs become
of interconnected better at predicting the next
layers that process word in a sequence
information
LLM Learning Method
LLM Scale
Developer LLM Year Released # of Parameters Corpus Size
Meta LLaMA 3 2024 150 billion 4 trillion tokens
Google Gemini 1 2024 1 trillion 5+ trillion tokens
OpenAI GPT-5** 2024 2 trillion+ 20+ trillion tokens
OpenAI GPT-4 2023 1 trillion+ 10+ trillion tokens
Google PaLM 2 2023 540 billion 1+ trillion tokens
Meta LLaMA 2 2023 70 billion 2 trillion tokens
Anthropic Claude 2 2023 Not publicly disclosed 10+ trillion tokens
OpenAI GPT-3.5 2022 175 billion 570 billion tokens
Google PaLM 2022 540 billion 1+ trillion tokens
Meta OPT-175B 2022 175 billion 180 billion tokens
* Parameter refers to the variables that the model learns during the training process
** Planned Release
What can LLMs
do?
• Generate different creative
Generate formats (text, images,
videos, audio)
• Answer questions in an
Answer informative way
• Translate languages,
Translate breaking down
communication barriers
• Help write different kinds of
Help content (emails, letters,
reports, reserachpapaers)
Key Players in the
GenAI Landscape
Top 5 Players and their Chatbots
Company Latest Model Chatbot
OpenAI GPT-4o ChatGPT
Google Gemini 1.5 Flash Gemini
Cohere Command R+ Coral
Anthropic Claude 3 Claude
Mistral Mistral le Chat
Limitation and
Challenges of
LLMs & GenAI
Data Bias: Limited Reasoning Explainability:
LLMs can reflect Ability: Understanding how
biases present in LLMs struggle with LLMs arrive at their
their training data tasks requiring outputs can be
(Examples: Gender complex reasoning or difficult
bias in job titles, logic
racial bias in news
articles)
The GenAI
Social Impact:
A Double-
Edged Sword
The Good
Language Translation: Personalized Learning: Accessibility Tools:
GenAI can translate GenAI tutors can GenAI can create tools like
languages on-the-fly, personalize learning for text-to-speech or speech-
bridging cultural gaps and each student, boosting to-text, promoting
dissolving language engagement and inclusivity and access to
barriers effectiveness. information for people
with disabilities.
Artistic Inspiration: Scientific Discovery: Product Development:
GenAI sifts through GenAI can analyze vast GenAI can assist in product
massive data, uncovering amounts of data to design by generating ideas
patterns to speed up identify patterns and and testing prototypes,
scientific discoveries. accelerate scientific leading to more innovative
breakthroughs. solutions.
The Bad
Deepfakes and Erosion of Trust: Biased Training Data:
Synthetic Media: Deepfakes can sow GenAI models are
GenAI can be misused doubt in legitimate trained on massive
to create realistic but information sources, datasets that may
fabricated videos or making it difficult to reflect existing societal
audio recordings discern truth from biases. These biases can
(deepfakes) that can fiction and eroding be perpetuated in the
spread misinformation trust in media and outputs generated by
and damage institutions. the model.
reputations.
Exacerbating Job Automation: The Need for Reskilling
Inequality: As GenAI becomes and Upskilling:
If GenAI is used in areas more sophisticated, it The changing nature of
like hiring or loan has the potential to work due to GenAI
applications, biased automate tasks necessitates investment
algorithms could currently performed by in reskilling and
unfairly discriminate humans, leading to job upskilling programs to
against certain groups displacement in certain ensure a smooth
sectors transition for the
workforce
The GenAI
Social Impact:
Will GenAI Steal
My Job?
GenAI excels at repetitive tasks:
Data entry, scheduling, basic
analysis are prime candidates for
automation
Human skills remain irreplaceable:
Creativity, critical thinking,
problem-solving, and social skills
are essential for many jobs.
Safe Career Choices
Does it require interacting with
people and using social
intelligence?
Does it involve creativity and
coming up with clever solutions?
Does it require working in an
unpredictable environment?
Tegmark, M. (2018). Life 3.0: Being human in the age of artificial intelligence (p. 121). Vintage Books
GenAI &
Chemical
Engineering: A
Perfect match
Chem. Engineering Use Cases
Appendix
30 Terms You Need To
Understand Artificial Intelligence:
Top 5 Considerations For Implementing
AI in Large Enterprises:
Read my Blog post here
The Best Way to Talk to an LLM:
Learn how to correctly prompt a GenAI
Chatbot
Essential Layperson Guide To How LLMs
Work:
Way Forward
• Professional Courses:
• Machine Learning in Chemical Engineering ( Dr. Zohaib Attiq Khan)
• Prompt Engineering
• Generative AI
• National Competition
• Prompt Engineering
