Matthew Turk: [ Home ] [ Research ] [ Teaching ] [ Personal pages ] [ Contact Info ]

Notes on Research

It's not the same as taking a course.

Courses provide a clear framework for how to succeed: read the syllabus, attend the classes, do the assigned reading, complete the homeworks, study for the tests, etc. - all with clear deadlines and boundaries of what's included and what's not included. Every problem has a specific answer. Generally, the basic mechanism is "learn this concept and apply it to this well-matched scenario."

Research is very different. Deadlines tend to be few and far between - e.g., a conference submission deadline or a final presentation or dissertation. There's no syllabus, and no single sources of all you need to know (such as lectures and textbooks). There's no answer in the back of the book that you're working toward. The basic mechanism is "Here's a (possibly not well-defined) problem - what do you need to learn to solve it?"

People who are great in coursework are not necessarily great at research, and vice versa. Research is a different skill, and it needs to be learned. That takes real work and dedication.

Choosing the problem is the most important step.

Articulating a clearly-defined research problem is vital for making real research progress. This is much more than just identifying the topic; a well-defined research problem includes a clear description of the task, why it is important and relevant, what the current state-of-the-art is, and how progress or success is to be measured.

Getting started

There are several ways to get started with research. All of these can be useful: taking relevant courses, going to relevant talks, reading about the topic (and related areas), reading research papers on various approaches to the specific problem, implementing a competing or state-of-the-art system to explore its behavior or performance. However, real research progress starts with trying something. Setting out to learning everything about the field or problem before you start doing the actual research is likely to cause significant and unnecessary delay. The more successful path is to learn a little, actually try something, analyze (and discuss with others) why what you did isn't adequate, learn a little more, modify your attenpted solution, analyze (and discuss with others) why it's still not adequate, etc. Rather, rinse, repeat. Research is an iterative process, and DOING SOMETHING needs to be an early (and frequent) component.

To a first approximation, only the results matter.

While the process of doing research involves lots of things that are educational and provide good experience for the researcher, at the end of the day the only thing that really matters is the result. E.g., do we know something now that was not known before? Do we have a method/algorithm/system that does something not done before, or in a manner (accuracy, speed, robustness, etc.) that is significantly better than prior methods/algorithms/systems? And can this be verified or proved?

Building it is not enough.

"I built a system to do X" is not enough. In research, you have to systematically assess/evaluate what you created. Unless your research is provably correct, this usually means an experimental assessment that measures performance along the important dimension(s), done in a scientifically valid way. Thinking through how to assess a system is something to be done before you build it, not as an afterthought. In some cases, such an assessment may be about accuracy or error rates or ROC curves; in other cases, it might involve a user study. Or both.

Your advisor

Interacting with your advisor(s) at every step of the research process is important. A student should not do a whole research project, or a large chunk of a research project, and then show it to his/her advisor. It is an interactive process, with frequent touching base on priorities, plans, problems, accomplishments, etc. Such meetings are opportunities to get useful feedback and should be welcomed. Don't wait for your advisor to ask for such a meeting - take the initiative to schedule it.

On the other hand, a student research should not expect to be told what to do by the advisor. A researcher needs to be proactive and "own the project," taking responsibilty for pursuing new independent ideas, for watching out for dead ends or unfruitful directions in the work, for figuring out the unknowns in the process. Other reserachers (students, visitors, faculty) should be viewed as great resources to discuss your work with; regularly talk with others about your (and their) work; critique each other's work, brainstorm, offer advice, ask for help. DON'T ISOLATE YOURSELF AND TREAT YOUR RESEARCH AS AN INDEPENDENT ISLAND.