Tips for Getting Started with Research

Starting research in AI, machine learning, or neuroscience can feel overwhelming given how fast these fields move. Here are some practical tips to help you get oriented and make progress.

Read Broadly Before Going Deep

Before diving into a specific problem, spend time surveying the landscape. Read survey papers and review articles to understand the major themes, open questions, and historical context of a field. This gives you a map before you navigate the territory.

• Use resources like Semantic Scholar, arXiv, or Google Scholar to find influential papers.
• Sort by citation count to identify foundational work.
• Look for recent survey papers (typically titled “A Survey of…” or “A Review of…”) to get a condensed overview of a subfield.

Pick a Concrete Problem Early

Broad curiosity is valuable, but vague goals stall progress. Narrow down to a concrete question or task as soon as you can — even if it changes later. A specific problem gives you something to read against, code toward, and discuss with others.

Reproduce Before You Innovate

One of the best ways to deeply understand a method is to re-implement it from scratch or reproduce a published result. This forces you to confront details that papers often gloss over, and it builds the hands-on intuition needed to extend or critique existing work.

Keep a Research Journal

Write down what you read, what you tried, and what you learned — even informally. A simple log of daily notes pays dividends when you need to retrace your thinking, write up your work, or explain your approach to a collaborator.

Engage with the Community

Research is a social activity. Engage with others to accelerate your learning:

• Attend lab meetings, reading groups, or seminar series.
• Follow researchers on social platforms or subscribe to their blogs.
• Ask questions — in person, in office hours, or in public forums like the ML Subreddit or Discord communities around specific tools and frameworks.

Learn to Read Papers Efficiently

You do not need to read every paper end-to-end. A common approach:

1. Abstract and introduction — decide if the paper is relevant.
2. Figures and results — understand what was done and what was found.
3. Methods — dig into the details if the work is directly relevant to yours.
4. Related work and conclusion — find additional papers to pursue.

Manage Scope and Avoid Rabbit Holes

It is easy to spend weeks reading without producing anything. Set time limits on literature surveys, and regularly ask: what is the minimum I need to understand to make progress on my problem? Ship small results early, even if they are negative.

Use Version Control from Day One

Treat your code and experiments like you treat your writing. Use git to track changes, and structure your project so that others (or future you) can reproduce your results from scratch.

Ask for Feedback Early and Often

Share rough drafts, incomplete code, and half-formed ideas with advisors or peers. Early feedback is cheaper than late-stage revisions. Do not wait until things are “ready” — the goal of feedback is to find out what still needs work.