Research Process

Steps in the process

• Define a problem - Define the problem you are trying to solve
• Read - Read to gain skills and to understand other attempts at your problem
• Explore - Try out the techniques, software, mathematics and possible solution ideas
• Propose - Write out a plan for attacking the problem
• Develop - Develop your solution
• Validate - Prove that you have a good solution
• Publish - Tell the rest of the world what you did so that they can benefit from your work.

Define a problem

A key to successful research lies in finding good problems. In this step you must address research questions 1 (what is the problem?) and 2 (who cares?). A good research problem should provide intellectual leverage. "If I succeed at solving this problem, then there will be many different benefits in many applications." If the solution only has a single application then generally it is product development rather than research. A good research problem has leverage.

A good research problem is also one where you or your lab have unique advantages over others. These may be specialized computing resources, mathematical or software techniques that have not been tried on this problem, a new insight into the nature or the problem or access to data or other resources.

Developing the problem

Start with a one sentence statement that clearly addresses research questions 1 (what is the problem?) and 2 (who cares?). Getting this sentence right will clarify your thought before investing a lot of work in more words. You will also have frequent need to give a short direct answer to "So what are you working on?" You should always have this answer. Review this sentence with others to see if it makes sense.

Expand your sentence into a paragraph of about 1/3 page that addresses the same questions. This will expand your thoughts again without a huge investment in words. If your thoughts will not fit in a paragraph then you do not yet understand your problem. Review this paragraph with others and see if they can answer questions 1 and 2 from your paragraph.

Read

The only way to answer research question 3 is to read a lot of books and papers. However, it is important to read efficiently. Reading every word of every paper is a waste of time. You need to read with a purpose. That purpose is defined by your research problem and by any technologies that you think might support your solution.

The first question when reading a paper is "How might this paper address one or more of the five research questions as they relate to my problem?". Knowing the answer to this question tells you what you are looking for in the paper. For example, if you are looking for a way to validate your results, you may want to skip immediately to the end of the paper to see how they validated their results. If you are looking for experimental data about some approach, skip to the results summary to see what their conclusions were. They may not have tested for what you are interested in. In which case you just saved an hour of reading. Read to answer your questions.

Take Notes.

Any paper worth reading is worth taking notes about. There is nothing more frustrating than to start writing a thesis or a paper, to remember a source that you read and to not remember where you found it or what the exact conclusions were. You read the most at the beginning of a project and you publish at the end. You forget a lot between those two.

Do not take notes on odd scraps of paper. Get a bound notebook that you keep with you always or open a file or web page for collecting all of your notes. Good notes include:

• A full bibliographic reference. This is easy to get when you read a paper and a pain to find again when you are trying to write your papers.
• How does this paper address the 5 research questions? Filter this by relevancy to your problem. When you characterize this work in your papers you will need answers to all five questions to help you differentiate this paper from other work and from yours.
• Lists of other references to search. When you find a good paper, it should lead you to other papers. There are two ways for following references.
  • The reference list of the paper will lead you to the work on which this paper was based. This will lead you to foundation concepts. Make sure they are relevant to your problem. There is too much in the world to read it all.
  • WWW search sites designed for scholarly work will frequently provide lists of other papers that cite this paper in their bibliography. This is very helpful in finding more recent work on a topic. You need to find the latest stuff
To find stuff to read you will need some keywords on which to search. Watch for when papers cross disciplinary boundaries. Different disciplines call the same concepts by different names. Keep lists of papers to read along with why you thought you should read them.

Good search sites for Computer Science

• ACM Digital Library available for free from any machine on BYU campus. This contains everything published by ACM plus some other stuff. Papers have all ACM citations listed.
• CiteSeer an index of scholarly computer science papers that people have published on the web. This also includes good citation links. Frequently helpful in finding papers from non-mainstream conferences.
• IEEE Digital Library - free from on campus. Search all IEEE publications.
• Google Scholar A variant of Google that targets scholarly papers. Behaves much like CiteSeer.
• Harold B. Lee library search of ACM, IEEE and other sources.

Explore

Before becoming fully committed to a project, it is very useful to do some exploration. Try simple prototypes of algorithms or techniques that you discovered while reading. Try simple projects with any new software that you have never used before. Try prototypes of some of your research ideas. You do not really understand a problem, technique or piece of software until you try to work with it. Set small goals (it helps to explore with a purpose) and see how easy or hard it will be to do the things you want.

Exploration is very important to determine if you really think there is fruit in this research approach. However, there is a trap of letting exploration consume all of your research time. There is also the trap stuffing your entire research plan into your exploration. This can spend a lot of time without a plan for what you will need in the end.

Any code you write, plan on throwing it away and starting over. The goal is to learn, not to develop usable code.

Propose

Writing is an important discipline to your thinking. Graduate programs require proposals because of the discipline it imposes on young researchers. Writing a proposal accomplishes two goals: 1) writing ideas down makes them concrete and forces you to sort out insights from wishful thinking, and 2) a proposal forces you to articulate your ideas to someone else. The end of the process is publication. If you can not write a proposal it is very doubtful that you have something worth publishing.

Proposal steps:

• Review the sentance and paragraph that you wrote while defining your problem. After reading and exploration you may need to rewrite these two items. Hopefully you have learned a lot.
• Expand your paragraph into 1 page. This page should address all five of the research questions. For question 4 on your solution you should sketch the key insights that you believe will lead to a solution.
• Look at the appropriate format for your proposal
  • MS Degree
  • Ph.D.
• Review the 5 questions and write an outline for how you will address them. Consider your format requirements and the guidelines in the next step.
• Write the proposal
  • 1/2 page answering question 1 (what is the problem?)
  • 1/2 page answering question 2 (who cares and why?)
  • 1 page on question 3 (what have others done?). A key part of any proposal defense is finding out if you know the literature in your field. This is where those notes that you wrote while reading come in very handy.
  • 1 page sketching out the key insights for question 4 (what is your solution?). At this point you are sketching ideas of how to attack the problem.
  • 1/2-1 page addressing question 5 (how will you test your solution?). It is very important that you show a plan for how you will identify success.

Develop

Validate

Publish