বাংলা

Goal guide · start from where you are

Thesis, research & publication

Learn to ask a defensible question, evaluate evidence and communicate a contribution. Aim for quality—not a promised world-class journal.

Find my starting pointGive me one action this week

First, a reality check

Know the route before the checklist

A strong thesis is a research process, not a paper-shaped document. In computing, respected conferences as well as journals can be relevant. Venue quality depends on the field and review standards, not just a metric. Acceptance is never guaranteed; honest negative results can still teach you a great deal.

First thesis / research apprenticeship

Start with a supervisor, a bounded question and a reproducible baseline. Learn methods before optimizing publication count.

Journal or conference submission

After evidence is solid, choose a venue whose scope matches the contribution. Read recent accepted work, ethics rules, author instructions and fees.

Not a test or a score

Where are you starting today?

Jump to the point that fits. If earlier evidence is missing, return to it. Nothing is saved or scored, and stages have no fixed completion deadline.

Stage 1 / 4

Find a question small enough to test

A fashionable topic is not yet a research question.

What to do

  • With a supervisor, choose a narrow area and read a small set of relevant papers. Record each question, method, data, result, limitation and open issue; verify every citation.
  • Write a testable question and a feasibility plan covering time, compute, data access and permissions. Get institutional ethics approval when required before collecting participant or sensitive data.

Enough for now when…

A short literature matrix and proposal with a question, baseline, evaluation plan and ethical constraints.

Not yet: Do not start with 'add AI to something' or collect sensitive data first and ask about consent later.

Reference for this stage: GOV.UK Service Manual

Stage 2 / 4

Reproduce, compare and test assumptions

You need to know what your result is being compared against.

What to do

  • Reproduce a baseline on a permitted dataset or controlled system. Record environment, versions, seeds where relevant, preprocessing and evaluation; separate development from final evaluation data.
  • Change one justified factor, compare fairly and inspect failures. Use field-appropriate statistics, uncertainty, ablations or qualitative analysis with your supervisor's guidance.

Enough for now when…

A reproducible baseline, documented comparison and limits on what the findings actually support.

Not yet: Do not chase only the best score, leak test data or hide negative runs and inconvenient observations.

Reference for this stage: ACM

Stage 3 / 4

Write, defend and choose a venue

A reader should be able to trace your claims back to evidence.

What to do

  • Write the problem, related work, method, evaluation, findings and limitations. Distinguish your contribution; agree on authorship with collaborators and preserve source attribution.
  • Rehearse a thesis defense with critical questions. For publication, inspect venue scope, editorial/review process, recent work, submission rules and costs with your supervisor; do not select by a flattering invitation.

Enough for now when…

A reviewed manuscript or thesis, an artifact package and a venue-fit explanation with honest limitations.

Not yet: No purchased authorship, fabricated citations, paper mills or simultaneous submissions where venue rules prohibit them.

Reference for this stage: ACM

Stage 4 / 4

Respond to review and make work usable

Research improves through scrutiny, not just acceptance emails.

What to do

  • Respond point by point, distinguishing changes, evidence and respectful disagreement. If rejected, diagnose fit and methodological gaps before choosing the next step with your supervisor.
  • Share code, data or materials only when consent, licenses and agreements permit. Provide setup and reproduction instructions; document what cannot be released and why.

Enough for now when…

A revision log, reproducible permitted materials and a clear explanation of remaining uncertainty.

Not yet: Do not erase failed experiments or overstate practical impact to make the conclusion exciting.

Reference for this stage: ACM

Keep your judgment switched on

Common traps & sensible AI use

  • A high metric alone does not prove a venue is suitable; inspect the field, real peer review and publication practices.
  • Publication costs, data permissions and compute requirements belong in the plan, not in a last-minute surprise.
  • One small study cannot support claims about everyone. Report sampling and evaluation limits explicitly.

AI can assist. You remain responsible.

AI is not evidence or an accountable author. Verify citations, code and analysis against originals; do not upload restricted data. Follow the current institution and venue rules for permitted AI use and disclosure, including research-method use.

Explore AI-resilient skills →

One small action is enough

Make this week count

Read three closely related papers, build a literature matrix and propose one reproducible baseline to a supervisor. A smaller question with honest evidence is a real start.

Afterwards, write three lines: what I did, what I learned, what I will change next. Today's direction is a draft—not a lifetime contract.

The skills underneath the goal

Pair this with a career roadmap

These are relevant starting points, not a list of prerequisites. Choose one; other careers may also fit your specific goal.

Adjacent directions

Build engineering evidence through your final-year project →

Sources & requirement checks

This roadmap is CSE-Pothik editorial guidance. Sources provide specific official references, not endorsement of the entire plan. Reading guidance may be free while applications, tests, tools or publication have costs. Rules change: recheck official pages before acting. External references are primarily in English.