Research Subject

By. Ankita Sharma

E11389
 “Who here wants to work in AI, Data
Science, or build a startup?”
 “Do you know all of these came from
research papers?”
Google, Open AI , Tesla
“All of these billion-dollar companies started
as research projects.
Today, you’ll see how YOU can think like a
researcher, even if you don’t want to become
a scientist.”
 How?
 How it started:
 In 1996, (PhD students at Stanford) were researching
how to rank web pages better.

 What happened next?

 It worked better than any existing search engine!
 They turned it into a product → Google → now a
trillion-dollar company.
 How Open AI started:
 OpenAI is a research company in Artificial
Intelligence (AI).
 Their goal was to make AI smart but safe
for humans.
 (They started by doing pure research –
publishing papers, testing models.
 They created GPT Generative Pre-trained
Transformer).
 This research turned into real products.
3. Tesla → Research in Electric Cars & AI
for self-driving

 Tesla invested in battery research,

renewable energy, and autonomous
driving algorithms.
 Self-driving cars came from research in
computer vision + AI + robotics.
 Now Tesla isn’t just a car company—it’s also
a research-driven AI + energy
company.
 Research = A smart way to find answers.
 It’s like training an ML model:
 You collect data → like gathering facts.
 You analyze patterns → like finding relationships
in features.
 You get predictions/insights → like generating
new knowledge.
 So, Research = Brain Training for Humans!
 How can I make an app?
 How do humans trust AI decisions?

Question : Which one feels like a project vs.

a research problem?
 Project → solving a known problem with
existing solutions

 Research → finding new knowledge or

improving existing ones
A Project
 Goal: Solve a problem using existing
knowledge or technology.
 You already know how to solve it, you
just need to build it.
 Example: Making a mobile app for
attendance tracking.
 We already know how to make apps.
You’re just applying known methods.
“Can AI write better code than
humans?”
Project way: Make an AI that generates
code.
Research way: Compare AI-generated
code vs human-written code for bugs,
efficiency, readability → Find out if AI
really outperforms humans.
Super Simple Analogy

 Project = Cooking a known recipe from

YouTube.
 Research = Inventing a new dish no one has
tasted before.
 UNIT-1
CHAPTER-1
Definition of research
 Definition of Research
 Research is a systematic process of
finding new knowledge, solving problems,
or improving existing solutions by
collecting and analysing information.

 It’s curiosity with a plan

 You ask a question → collect data →
analyse → find answers
Research Example: “Does listening to classical
music improve concentration while studying?”

1.) Questions/observations and knowledge

 You notice that some people say classical music
helps them focus.
 You also read some articles suggesting music may
affect brain activity.
 This sparks curiosity → Can music improve
concentration?
2. Develop the research topic

 The broad topic becomes:

“The impact of classical music on

concentration and focus.”
3. Develop the research question

 “Does listening to classical music

improve concentration during a 30-
minute study session compared to
studying in silence?”
4. Choose a study design
 Experimental design: Two groups of
students—one listens to classical music, the
other studies in silence.
 Measure their concentration using a reading
comprehension test after studying.
5. Data collection

 Recruit 20 students.
 Randomly assign 10 to the “music group”
and 10 to the “silence group.”
 After 30 minutes of studying, give all
students the same test.
 Record their scores.
 6. Data analysis
 Compare the test scores of both groups.
7. Draw conclusions
 If the music group scored higher →
classical music may improve
concentration.
 If no difference → music might not affect
focus, or more research is needed.
8. Inform others/application in
practice
 Publish your findings, present them in a
school or academic journal, or share
with students and teachers.

 Others can apply your findings to

improve study habits.
Objectives of research
 Objectives of Research

 To discover new facts

 To verify old facts
 To solve problems
 To develop new → tools, methods, or
theories
 To improve existing solutions
Objective What it means ML Example

Find something completely Discovering that people use a

To discover new facts new that no one noticed new slang word in social
before. media sentiment analysis.

Verifying that “more training

Check if what we already data improves model
To verify old facts believe is still true. accuracy” using a new
dataset.

Use research to fix a real Building a simple spam email

To solve problems issue. classifier to reduce junk mail.

Developing a new
To develop new tools, Create something new to
preprocessing method for
methods, or theories make tasks easier or better.
cleaning noisy text data.

Make an existing idea, Improving a chatbot by fine-

To improve existing
method, or system work tuning a pre-trained model for
solutions better. better answers.
Types of Research
 “Imagine you’re helping Amazon improve its review system. First, you
learn how NLP models understand language (Basic). Then you build a
tool to detect bad reviews (Applied). Before that, you just explore the
dataset to see common words (Exploratory). You summarize how many
are positive/negative (Descriptive). You experiment with different
models to see which works best (Experimental). You manually read
some reviews to see WHY people are upset (Qualitative), and finally,
you measure the model’s accuracy with numbers (Quantitative).”
 Basic Research → Learn the math behind
recommendation algorithms like collaborative
filtering.

 Applied Research → Actually build a recommender

that suggests videos to users.

 Exploratory Research → Look at YouTube watch

history to find hidden viewing patterns (e.g., people
who watch cooking videos also watch travel vlogs).
 Descriptive Research → Describe what categories
are most watched (e.g., 40% music, 30% gaming).

 Experimental Research → Test two

recommendation models (content-based vs.
collaborative) and see which gets more clicks.

 Qualitative Research → Interview users about why

they like or dislike the recommendations.

 Quantitative Research → Collect click-through

rates and view counts to measure success.
 Machine Learning
Type of Research Simple Meaning Everyday Life Example
Example

Studying how gradient Learning how the brain

Learning just for knowledge, no
Basic Research descent or transformers stores memory without
direct product
work mathematically making a product

Creating a memory-
Using knowledge to solve a real- Building a spam detector
Applied Research improvement app based on
world problem or crop disease model
neuroscience
Clustering a new TikTok
Exploratory Exploring when you don’t know Visiting a new city and
dataset to find hidden
Research much yet exploring unknown areas
video categories
Reporting 80% of reviews
Descriptive Describing what exists, no “why” Writing a blog describing
are positive, or user
Research or prediction what you saw in a trip
demographics
Trying different learning Testing two study methods
Experimental Testing cause-and-effect by
rates to see which to see which improves
Research changing one thing
improves accuracy exam scores

Reading user feedback

Qualitative Focusing on opinions & feelings Asking students how they
comments to see why
Research (non-numerical) feel about online classes
they like a chatbot
Calculating accuracy, Counting how many
Quantitative Focusing on numbers &
precision, recall of a students attend online vs.
Research measurements
model offline classes
 Other Domains You Could Use
 Chatbots → Understanding how ChatGPT works (basic), building a
customer support bot (applied), testing different prompts
(experimental).
 Fake News Detection → Exploring headlines (exploratory),
summarizing % of fake news (descriptive), testing models
(experimental).
 YouTube Recommender → Describing user watch history (descriptive),
testing recommendation algorithms (experimental).
 Why is Research Important in Computer
Science?
 Taking the example of Cybersecurity

 Helps discover new threats

 Improves security systems
 Finds weaknesses in existing systems
 Builds smarter defense tools
 Protects privacy of users
 Keeps up with evolving hackers
 Creates security awareness
Helps us learn new things
 Example: Research in NLP created models like ChatGPT.
Solves real-life problems
 Example: AI research helps detect fake news or predict
diseases.
Improves existing technology
 Example: Research made deep learning models faster and
more accurate.
Creates new innovations
 Example: Research in computer vision led to self-driving cars.
Makes apps and tools better for users
 Example: Netflix recommendations got better through
research.
Finds problems and fixes them
 Example: Research found AI bias and improved
fairness.
Opens career and startup opportunities
 Example: AI research created new jobs like Data
Scientists and AI Engineers.
3. Research Process
 Definition: The sequence of steps researchers follow to conduct a
study, often aligned with the scientific method but broader in scope.
Typical Steps:
 Identify a Problem or Topic
 Review Literature
 Formulate Research Questions or Hypotheses
 Choose Methodology (Qual, Quant, or Mixed)
 Design Research Tools (surveys, interviews, etc.)
 Collect Data
 Analyse Data
 Interpret Results
 Draw Conclusions and Implications
 Disseminate Findings
Hypothesis Testing

 Hypothesis testing compares two opposite ideas about

a group of people or things and uses data from a
small part of that group (a sample) to decide which
idea is more likely true. We collect and study the
sample data to check if the claim is correct.
Does a New Teaching Method Improve Math
Scores?
 A school wants to test if a new teaching method improves
students’ math scores and increases the percentage of
students who pass the final exam.
 We'll perform two tests:
 A mean test using μ (mu)
 A proportion test using p
 Step 1: Hypotheses

 Null hypothesis (H₀): μ = 70

 Alternative hypothesis (H₁): μ > 70
 Null hypothesis (H₀): The average score is still
70. (No improvement)
 Alternative hypothesis (H₁): The average
score is now higher than 70. (Improvement)
 2. Testing the Proportion (p – Population
Proportion)

 The school also wants to know if more students are

passing the final exam than last year.
 Previously, 60% of students passed.
 This year, they check the percentage of passing
students in a sample.
 Hypotheses:
 Null hypothesis (H₀): The pass rate is still 60%.
 Alternative hypothesis (H₁): The pass rate is now
higher than 60%.
Parameter What is being tested? What decision is being made?

If average student scores Is the new method improving

μ
increased how well students perform?

Is the new method helping more

p If more students are passing
students succeed?
Why Do We Need Hypothesis Testing?
 Hypothesis testing is used to make decisions based on
data rather than guesses or opinions. It helps us
determine whether a claim about a group (like people,
products, or results) is likely to be true or false based
on evidence.

Imagine This:
 You hear a claim like:
 “This new medicine works better.”
“Students are scoring higher this year.”
“More than half the customers are happy with the new
product.”
 Why They Use Hypothesis
Field Example Organizations
Testing

Healthcare Pfizer, WHO, hospitals Test treatments or drugs

Education Schools, colleges, education boards Evaluate teaching effectiveness

Business Amazon, Google, startups Test products, ads, pricing strategies

Government CDC, Census Bureau, ministries Test public policies or trends

Science NASA, research labs, universities Validate experiments

Manufacturing Toyota, GE, quality control teams Improve product quality

Sports Olympic committees, teams, trainers Test training effects or equipment

 Hypothesis testing example
 Example: Average Weight of Apples
 A fruit seller claims that the average weight of apples = 100 grams.
A student wants to test if the apples are actually lighter.
 Steps in Hypothesis Testing:
 State Hypotheses
 H₀ (Null): Apples weigh 100 g on average.
 H₁ (Alternative): Apples weigh less than 100 g.
 Collect Data
The student picks 5 apples and weighs them:
95 g, 97 g, 98 g, 94 g, 96 g.
→ Average = 96 g.
 Compare with Claim
 Claimed mean = 100 g.
 Sample mean = 96 g → noticeably smaller.
 Decision
Since the sample mean (96 g) is much less than 100 g,
we reject H₀ and say apples are lighter than claimed.
 Logic of Hypothesis Testing
 Start by assuming the null hypothesis (H₀) is true.
 Collect sample data and calculate a test statistic.
 Compare the test result with a critical value or p-value.
 If the evidence is strong enough, reject H₀ in favor of the
alternative hypothesis (H₁).
 If not, we fail to reject H₀ (but never fully “prove” it true).
Types of Errors
 When making a decision, two kinds of errors may occur:
 Type I Error (False Positive)
 Rejecting H₀ when it is actually true.
 Example: Concluding a new algorithm is faster when it’s not.
 Probability of Type I error = α (significance level).
 Type II Error (False Negative)
 Failing to reject H₀ when H₁ is true.
 Example: Concluding a new algorithm is not faster when it actually is.
 Probability of Type II error = β.
 Power of a test = (1 – β).
 Significance Level (α)
 The threshold probability of making a Type I error.
 Common choices: 0.05 (5%), 0.01 (1%).
 Example: If α = 0.05 → We are willing to risk a 5%
chance of wrongly rejecting H₀.
 p-Value
 The probability of obtaining the observed data (or
something more extreme) if H₀ is true.
 If p ≤ α → Reject H₀.
 If p > α → Fail to reject H₀.
 Research Planning and Flowcharting

1. Planning a Research Study.

a) Scope
 Defines what the research will cover.
 Outlines boundaries (what is included & excluded).
 Example: If studying “AI in Healthcare,” scope may cover diagnosis
systems, but exclude surgical robotics.
 b) Limitations
 Acknowledge constraints in the research.
 Can be due to:
 Time (short research duration)
 Resources (limited budget, tools)
 Data access (restricted datasets)
 Methodology (cannot test all variables)
Timeline
 Break the study into phases (proposal → literature review → data
collection → analysis → reporting).
 Represent with Gantt charts or milestones.
 Helps in monitoring progress & deadlines.
2. Flowcharts for Research &
Algorithms
 Flowcharts = visual diagrams showing logical steps in a
process.
 Algorithmic Representation:
 Shows sequence of steps in program logic (input → process
→ output).
 Example: Flowchart for searching/sorting algorithm.
 Research Planning Representation:
 Maps the research workflow (problem identification →
hypothesis → methodology → data collection → analysis →
conclusion).
 3. Tools for Flowcharting
 Lucidchart – Professional, collaborative, cloud-based.
 Draw.io – Free, simple, integrated with Google Drive.
 Canva – User-friendly, visually appealing, good for presentations.
 4. Best Practices for Visual Representations
 Use standard flowchart symbols:
 Oval → Start/End
 Rectangle → Process
 Diamond → Decision
 Parallelogram → Input/Output