Research Tips

November 16th, 2021

A list of research tips for grad students.

---

Here are some research tips I’ve collected during the my past few years in grad school. Some are advice I’ve received in person, some I’ve read online and found really helpful, and some I’ve learned from my own personal experience.

Of course with any advice its applicability varies across different people and contexts. So, while I hope these tips can help you in your research, do take them with a grain of salt.

Choosing Problems and Ideas

• Always work on what you believe to be the most important problem.

  Rate problems by a combination of novelty, usefulness, and practicality.

• Be attached to problems, less attached on assumptions, and less attached to solutions.

• Combat paralysis by first believing that progress in the field is guaranteed.

  By paralysis I mean the feeling that there is nothing to be done with a given problem.

  Of course actual progress is not guaranteed, but it’s hard to make progress without first believing that there is progress to be made.

  It’s also easier to think “backwards” - given problem X is solved, what does that look like? What were the key advances that led to its solution?

• Maintain a list of unsolved problems that you think are important.

  Check this list occasionally when you acquire new information (e.g. new datasets, algorithms, hardware) to see if any can now be solved.

• Have a long term research and application vision.

  Where do you think your field is going in 5 years? 10 years? Where do you think it should go? What are the key applications and bottlenecks?

  These predictions don’t have to be accurate, but going through the exercise can help you build a deeper understanding of the key research problems.

• Incremental work is fine as long as they bring about new understandings.

  It’s also a great way to build confidence and learn about a field. Nothing beats personal “on-policy” experience of trying to solve a problem.

• Most ideas are bad ideas. Don’t be stingy with ideas.

• Test and discard new ideas quickly but carefully.

  Figure out and record what constraints or assumptions prevented them from working. This is useful in case these conditions change in the future.

• Don’t be afraid to make ideas concrete.

  It’s easier to show how concrete ideas don’t work as opposed to abstract ideas, so sometimes we might be reluctant to make ideas concrete in fear of discovering that they don’t work. It’s important to not procrastinate on this to make real progress.

• Just because a toy problem can be interpolated into the actual problem doesn’t mean the toy solution can be interpolated into the actual solution.

  Take care when designing the toy problem to make sure its experiments, solutions, and insights can be transferred to the real problem.

Staying Productive

• Trust your future self.

  Working on long-term projects with many parts can feel overwhelming. Trusting that your future self can fulfill their parts allow you to focus on the task at hand and not feel too stressed about all the things you have to do in the future.

• Ruthlessly prioritize.

  There are too many things to try in research and not enough time to try them. Always evaluate your options and focus on what you believe is the best use of your time.

• Be intentional about exploration.

  In research it’s very easy to meander about in a “sea” of ideas and come out the other side without getting anything done. To avoid this we should be aware of when we’re exploring ideas and document the process as we go.

• Group efforts are more effective for implementation than exploration.

  The former has clear objectives that everyone is aware of, so people don’t have to communicate and coordinate with each other as much.

  The latter typically requires quick iterations and series of small experiments, which can be executed much faster with a small number of people.

• Do pre-mortems.

  Imagine the current project fails in the future and list out reasons that explain its failure. Then try to avoid these reasons.

Staying Positive

• When reacting to new ideas, avoid “it won’t work because of X” and instead go for “it might work if X happens”.

  This “positive” framing is super important both when we discuss research with others and when we think about research ourselves. It creates a sense of forward momentum and focuses our attention on what’s possible.

• Don’t compare your day-to-day with an aggregate of everyone’s best days.

  This applies to things like scrolling Twitter feeds and browsing conference publications.

• Just because something is easy to you doesn’t mean it’s easy to everyone.

  A lot of folks, myself included, tend to put a premium on difficult and complex problems, ideas, and skills. However, using difficulty to gauge value is flawed since it may vary wildly among different people, and in any case value does not always correlate with scarcity. We should avoid this tendency and instead take advantage of the things that come easy to us.

Preparing Presentations

• Use the 3 R’s when making slides: Reduce, Reuse, Recycle.

  Reduce means using less slides and less content on each slide to make the point.

  Reuse means reusing previously made slides whenever appropriate.

  Recycle means not starting from scratch whenever possible - take figures, bullet points, tables from previous decks.

• Directly explain points of confusion.

  For example, using a keyword in multiple contexts with different meanings can be confusing. In addition to making the contexts clearer, directly explain the different contexts and definitions beforehand.

Misc

• Learn from new works not just their specific ideas but also their new ways of thinking.

• Strength of belief only weakly correlates with its validity.

  Don’t solely evaluate the validity of beliefs based on how strongly others believe in them or how popular the beliefs are. Evaluate them based on concrete evidence and from first principles.