Techniques Used To Do Research

 · 

so till now I have two techniques used for research : mathematical validation of a solution and creating a toy problem. What are the other techniques? :

  • pattern recognition
  • problem solving
  • toy problem
  • mathematical proof of solution validation
  • ablation studies
  • failure cases
  • thought experiments : what if book
  • cross-subject theory experimentation like game theory to multimodality

It made me think about the different phases of research: :

  • initial curiosity phase : where you try to think about the problem from a perspective of newbie
  • searching phase : where you try to find different paradigms that give you perspective to look at the problem like my multimodality perspective to see the audio visual problem and I think its the most basic one.
  • courage phase : where you pick an idea or paper and try to understand the problem from that tiny perspective. thats where courage is required because you dont know if it will work or not

is there any video or blog or linkedin post available? :

Yes. If your goal is to learn how researchers construct evidence and validate ideas, there are some excellent resources. I’ll group them by what you’re trying to learn.

1. How researchers think about evidence and experiments

Richard Hamming — “You and Your Research”

This is probably the single best talk for understanding how great researchers think.

You and Your Research (Bell Labs transcript)

Don’t read it as productivity advice. Read it as:

How do researchers decide what is worth investigating?

Michael Nielsen

His essays are excellent for understanding scientific thinking.

Michael Nielsen Essays

Particularly useful because he often discusses:

  • experiments

  • theory

  • research intuition

  • scientific discovery

2. Toy problems and mechanistic understanding

Anthropic: Toy Models of Superposition

This is one of the clearest examples of:

Big Question
↓
Tiny Toy Problem
↓
Mechanistic Understanding

While it is about neural network interpretability, the methodology applies to any research area.

3. How to design experiments

Andrej Karpathy

His talks are filled with practical experimental thinking.

Andrej Karpathy YouTube Channel

Pay attention to:

  • how hypotheses are formed

  • why ablations matter

  • how experiments answer questions

Yann LeCun interviews

LeCun frequently discusses:

Hypothesis
↓
Experiment
↓
Revision

rather than simply building models.

Yann LeCun Official Site

4. Research methodology for ML

The Gradient

One of the best blogs for understanding:

  • why papers are convincing

  • how to evaluate claims

  • experimental design

  • benchmarking

Search there for:

  • ablations

  • evaluation

  • benchmarking

  • causal reasoning

5. The framework you’re actually looking for

From our conversations, I think you’re gradually discovering that researchers use several different “evidence generators”:

Mathematical proof
Toy problem
Ablation
Benchmark
Intervention
Visualization
Failure analysis
Representation analysis

A useful exercise when reading any paper is:

Research Evidence Audit

Write:

What claim is being made?

How is the claim supported?

□ Theory
□ Toy Problem
□ Benchmark
□ Ablation
□ Visualization
□ Failure Cases
□ Intervention
□ Representation Analysis

After doing this for 20–30 papers, you’ll start seeing recurring patterns.

For someone with your interests (representation learning, audio-visual learning, temporal modeling, action localization), I would start with:

  1. Richard Hamming’s “You and Your Research”

  2. Anthropic’s Toy Models of Superposition

  3. The Gradient

Together they cover:

  • how to choose questions,

  • how to isolate mechanisms,

  • and how to build evidence for claims.

any books? :

Yes. But based on what you’ve been asking lately, I would not recommend books on productivity or “how to do research” first.

You’re trying to understand:

  • How researchers discover problems

  • How they build evidence

  • How they trust ideas enough to test them

  • How they move from intuition → hypothesis → experiment

For that, I’d recommend these books.

1. The Art of Doing Science and Engineering

If I could recommend only one book to a new researcher, it would be this.

Hamming discusses:

  • How great scientists think

  • How to identify important problems

  • Why some people produce more research than others

  • How to cultivate insight

What you’ll like:

He focuses on problem selection and scientific thinking, not productivity hacks.

2. The Craft of Research

This book teaches:

Claim
↓
Evidence
↓
Argument

which is essentially what a research paper is.

Useful because you’re beginning to think about:

How do researchers convince others?

This book directly addresses that.

3. How to Solve It

One of the most influential books on problem solving ever written.

Pólya’s framework:

Understand the problem
↓
Devise a plan
↓
Carry out the plan
↓
Look back

appears everywhere in science.

This is especially relevant because you’re trying to become more problem-oriented.

4. The Structure of Scientific Revolutions

This is not a practical research book.

It’s about:

How scientific ideas evolve.

You’ll learn:

  • paradigms

  • anomalies

  • scientific revolutions

Many research ideas come from understanding where current paradigms fail.

5. Thinking, Fast and Slow

Since you’ve asked about:

  • self-doubt

  • intuition

  • System 1 vs System 2

this book will help you understand:

  • cognitive biases

  • decision making

  • uncertainty

However, don’t read it as a research methodology book.

Read it as:

How my mind works while doing research.

6. The Beginning of Infinity

This is my wildcard recommendation for you.

The central idea is:

Progress comes from creating better explanations.

Not:

more data

Not:

more confidence

But:

better explanations

This is surprisingly close to how good researchers think.

For your current stage

Given everything we’ve discussed over the last year, I’d read them in this order:

First

The Art of Doing Science and Engineering

because you’re struggling with:

  • problem selection

  • confidence

  • research direction

Second

How to Solve It

because you’re trying to build a framework for approaching research problems.

Third

The Craft of Research

because you’re beginning to analyze papers in terms of evidence and argument.

Fourth

The Beginning of Infinity

because it helps develop the mindset:

My job is not to find certainty.

My job is to create and test explanations.

And honestly, that mindset would directly address many of the questions you’ve been asking recently about self-doubt, adaptability, and committing to research directions despite uncertainty.

Tags: