This post is just to update the tips and lessons I learned on research practices to remind myself and avoid future detours. Probably it generalizes to others as well. Basically, the research process will be reflected in the publication hence each of the tips I summarized below could be mapped into a specific segment of the paper.

Many Ideas Out There

  • Ideas could come from …
    • Take advantage of the advisor oracle - he/she has more experiences and better tastes typically.
    • Develop the habit to read best papers in the top conferences deeply. Understand the essence of the work and think about the limitations.
    • Discuss with peers, engineers during conferences, interns, daily lifes …
    • Pain or dreams in real-world life. Interestingly, fundamental and impactful works typically get inspirations from the real world complaint that people neglect.
  • Don’t forget that there should be a filter of ideas …
    • Do literature review; make sure the idea is valid (it will also be part of the related work and it is better to do it a priori).
    • Evaluate, discuss carefully before turning the idea into concrete research problem based on your own flavor.
  • Capture the ideas with written notes
    • If a great idea or inspiration dawns on you, note it down instead of letting it pass unconsciously.

Use Assisting Materials Well

  • Avoid getting drown in the sea of literature, textbooks. Directly pick the contents that will be helpful to the ongoing project or cater to your interest.
  • Greedy collection of all materials that are relevant is a bad idea. Learn to give up those that are less connected and just focus on those with real insights that would be helpful to the ongoing projects.

Try to Make Progress Daily

  • Don’t try to answer all questions same time, decompose the problem into measurable steps.
  • What is the next step: Research process is similar to SGD: take a number of baby steps aiming to reach an “optimal” solution. However, taking each step will cost probably significant time and efforts, and one should not brute force each possible solution without organization. Each step should be driven by the problem or sub-problems. Try to leverage on prior experiences and heuristics (yours, your advisor, collaborators…) to prioritize the most promising solutions and therefore reduce the search space.
  • Avoid blind engineering hack, it is helpful to ground the practices on heuristics, intuitions, theory.
  • Have a task stack for each project containing concrete next steps.
  • Zoom in and zoom out and crystallize the main line. Adjust the tasks in the stack if needed.
  • Try to pop the task daily.
  • Do not leap, take gradual steps one by one. It will help better to diagnose, fallback. If you leap too far and you are lucky, it is fine; however, if not, one has to start from the origin which suffers.
  • Have target milestones in mind: It is better to always target at a stable milestone, i.e., some concrete output in terms of implementation, theoretical justification of an argument, experimental result, rather than wondering before finding the “optimal” optimal solution without giving any visual outputs to be interpreted by others. I used to have a mindset that things have to be perfect before actually delivering it - it is often wrong since then one would just be trapped and get no products to discuss with others for perfection. It is totally fine to have an intermediate and clear unit of output with flaws and limitations since they will anyway be a stepping stone towards the final fancy output.

Log Down Stable Sub-conclusions

  • Log down the stuff (over Slack, Trello…) in self-organized units. Don’t get frustrated with detours, these all would go as part of the discussion in the paper, the most important thing is to conduct each step with thoughts and predictions, instead of blind search.

Time Management

  • One could never stress the importance of TM too much. It is the key to getting things done and alleviating stress.
  • Have a centralized calendar which denotes in detail ALL the things you need to deal with along the time axis. It will help give a better overview and gain control of your life.