How to do research well? A summary of suggestions by outstanding researchers

Choosing a research topic

  • 3 R criteria (JIAWEI HAN, UIUC)
    • Real data
    • Real but not solved problem (also in [1])
    • Real solution
  • Do not follow trends
  • Particular subproblem that not everyone is working on (SEBASTIAN RUDER, INSIGHT)
  • Prefer tasks with open source implementation
  • Become the strongest person in that specific problem  [2]
  • Match the strength & weakness of your environment [1]

Doing Research (CHENGLIN LIU, Chinese Academy of Sciences)

  • Derive more sub-problems based on application needs
  • Every step can be a top-tier venue
  • Read a lot by finding the relationships between different methods, and use other methods to solve yours
  • Reading & experiment simultaneously
  • Writing early & improve experiment

Innovation Approaches (CHENGLIN LIU, Chinese Academy of Sciences)

  • Improve current methods
  • Change data representation
  • Change problem definition
  • Use other methods to solve your domain problem
  • Combine different methods to improve

[1] HOW TO HAVE A BAD CAREER IN RESEARCH, DAVID PATTERSON, BERKELEY
[2] Q&A WITH YOSHUA BENGIO, U. Montreal