Research Methods

Today, I had an insightful conversation with someone who had an interesting viewpoint concerning the differences in the published work of schools in the east (e.g. GA Tech) and schools in the west coasts of the US. The basic idea behind his viewpoint was that schools in the east tend to approach scientific problems more as engineers than as scientist. More specifically, when attacking a problem, schools in the east tend to spend less time on defining the problem they are attacking and more time engineering a solution.

While I do not have the expertise to agree or disagree with his statement, the important take away from the conversation was that the most important step in the scientific process is to define the problem. I guess he is saying the investigation of a well defined problem is the difference between scientist and engineers. Engineering focus on the result while scientist focus on the process.

Later , I realized that the day before I had a conversation with a PhD student who proved his point. When I asked him how he went about publishing papers he told me he does not really start off with that as a goal. Instead, he said as a PhD student he is given projects to work on and if a paper is produced out of it then that is a plus. His words definitely did not agree with my spirit. What I would have liked to hear is - "I am given problems and I work with my adviser to understand and formally define the problem. Next, I take time and perform a literature review on the work in the problem's area. Following those steps, I develop algorithms and run experiments on them and compile the results. I can publish anytime after the problem is defined."

Below is a list version of how to attack attach research problems - this is obviously not the only method.

1. Find the problem --> Finding the problem may occur in one of three ways. First, after reading papers we may find facets of other's research which is untouched or difficult, and still has room to improve the state of the art. Secondly, we may see a practical problem occuring or notice something that is inefficient, whether that be through the newspaper, tv, friends, colleagues, etc. Lastly, because computer science has tentacles which may reach towards other areas, it may be possible to map the fields terminology and processes to the cs domain (an example of this is the general area of "Neural Networks").
2. Fully define the problem and its domain --> This step is of the utmost importance and must be completed before moving any further. Fully define the problem until it is well defined, sorted out, and ready to be attacked. This prevents people from questioning the foundations your research. A bonus of this step is that it allows you to do an elevator pitch to friends and colleagues at conferences.
   
3. Related Research --> Don't reinvent the wheel and stand on the shoulders of giants. Find and record all related work.
4. Develop Solution --> Figure out the best solution then design experiments to test solutions. When producing algorithms your foundations should be secure so you credibility is not taken.
   
5. Publish --> Publish all publishable work.
   
6. Rinse and Repeat