How to do research |
How to do research? And Related Issues of Ph.D Education "Can you tell me how to do research?", "Please tell me what topics should I pick to do my ph.d.thesis? " are two questions I periodically receive from young researchers and students in China. There are two difficulties associated with answering these two questions. 1. The first question is open-ended. To answer it honestly and accurately will require several hours of explanation. In fact, the process could take several years of face-to-face interaction between an advisor and a ph.d student. This is really what a ph.d. education involves. It is not the number of courses taken or the number of papers published. But something about the quality of and taste in research. I'll have more to say about this point below in connection with trying to answer this first question here . 2. Behind the second question lie a unintentional misconception by many young workers about "research" that probably has its origin in the martial arts novels of Chinese literature (武侠小说). In typical novels of this genre, the hero/heroine learns powerful martial arts technique through some secret magical instruction passed down from a master or a long lost instruction book. As a result, overnight s/he was able to defeat his/her enemy and achieve success. Thus, many students are constantly looking for that "magical formula" for research. Another common criticism of Chinese university education by Westerners is that most students learn to answer questions or pass exams extremely well, but there is little training in posing questions and formulating problems which are really the important parts and more than half of research. Growing up in such environment, it is natural for Chinese students to seek the magical formula for research success. It is also my purpose to address this issue here To give an answer to the first question, you can approach it at three levels. At the first level, your aim is to be that one scientific person in a century, e.g., the discoverer of the theory of relativity. I have no experience or advice to give at that level. All I can say is "good luck". At the next level, you want to be the founder of a research topic or make the breakthrough on a difficult problem. Again, there is no magic here. For otherwise, breakthrough will become commonplace. However, I do have one suggestion that worked well for me and which is generally applicable. This I offer below: I participated in a panel discussion 8 years ago which involves my answers to several general questions posed to me at the panel and which still apply today. ( The 1999 IEEE Conference on Decision and Control panel on "At the Gate of the Millennium – Are we in control?" Published in IEEE Control Systems Magazine Vol.20, #1, Feb, 2000) BEGINNING OF QUOTE 1. What are the 5 most notable research results in systems and control theory in the past 100 years? The "test of time" and traditions of history rule out mentioning anything developed in the past 25 years or involving living persons. Furthermore, scientific discovery often is a matter of standing on the shoulder of others. To single out specific results do not seem to be fair to others who laid the foundation. Instead, I propose to list couple of ideas that seems to me have influenced the development of our field in a major way 1. The fundamental role of probability and stochastic process in system work. 2. The concept of what constitutes a solution to a problem, namely, that which can be reduced to another routinely solved problem such as numerical integration 3. The notion of dynamics and feedback in all their ramifications The first item represents how knowledge from outside the field influenced our research while the third states what specific concepts our field contributed to other fields. The second item deals with how practices in science and mathematics are changed by technology. These notions are generic and have parallel in other topics and fields. 2. What are future research milestones (for the next 5-10 years) which will have a most significant impact on the field? Comment on what we should be doing now and in the near future to make such accomplishments possible. Also discuss your thoughts on: i.Education issues. How should we be training our students to meet the future challenges? ii.Technology issues. What we see coming up that will change the control landscape. iii.Computational tools. Scientific crystal balling has a notorious record in the past. The dust heap of past predictions is filled with gross miscalculations and estimations by noted scientists with the best of intentions. Let me try to approach the question "what's next in control systems in the 21st century?" in a somewhat different way. During my travel and lectures, I am often asked by young scientist/engineers starting out in their careers on what are profitable avenues of research to pursue. One is often tempted to point to ones own current research topic, which by definition must be the most interesting things to do. However, this is selfish and dangerous advice. My considered advice, which I myself have followed, is this: "GO FIND A REAL WORLD PROBLEM THAT A GROUP OF PEOPLE IS EAGER TO SOLVE, THAT HAPPENS TO INTEREST YOU FOR WHATEVER REASON, AND THAT YOU DON'T KNOW MUCH ABOUT. MAKE A COMMITMENT TO SOLVE IT BUT NOT A COMMITMENT TO USE TOOLS WITH WHICH YOU HAPPEN TO BE FAMILIAR" Such an approach has several immediate advantages. First, if you are successful then you have some free built-in Public Relations. Unsolicited testimonial by others is the best kind of publicity for your work. Second, most probably you have discovered something new or have found a new application of existing knowledge. In either case, you can try to generalize such discovery later into a fruitful research area for which you will be credited with its founding. Third, in a new problem area there is generally less legacy literature you will have to learn and reference. Fourth, a new problem area is like a newly discovered mine. For the same effort you can pick up more nuggets lying near the surface than digging deep in a well-worked out mineshaft. By the same reasoning, the probability of serendipity at work is also by definition higher in a new area. Lastly, even if you are unsuccessful in solving the original problem, you will have at the very minimum learned something new which will increase the chance of your success in future tries. My own personal experiences whether it was differential games, manufacturing automation, perturbation analysis in discrete event simulation, or ordinal optimization reinforces the above belief. Above all, faith in the ability of the future generation of scientists and engineers makes me an optimist in saying, "the best is yet to be, you ain't seen nothing yet". It is fine to make predictions and to look forward, but there is no need to get too obsessed with divining the future. As far as technology and computational issues are concerned, I believe I have already given my answer in the recent op-ed piece published in the June '99 issue of the IEEE Control System Magazine "The No Free Lunch Theorem and the Human-Machine Interface". I shall simply add by repeating what I said at my Bellamn Award acceptance "the subject of control which is based on mathematics, enabled by computers, is about to have a new birth of freedom under computational intelligence". END OF QUOTE. I now wish to elaborate on the last sentence in the above advice about making a commitment. In trying to solve a problem, one is always tempted to use tools with which one is familiar. This is very natural. Our training in school for the most part deals with tools. Exercises associated with such learning are always designed to fit these tools. As a result when we tackle a problem outside the textbook, our first instinct is to reach for these tools. But as engineers who ultimately must solve real world problems or as academic engineers who aspire to pioneer breakthrough research, such an approach is often not the best way. Basically, you are "a solution in search of a problem to fit". Too often, we bend the problem to fit the solution. Instead, it has been my experience that we should take the problem as it presents itself and not form any pre-conceived idea on how to solve it. Confronting a problem on its terms promotes looking at it in the simplest terms since we have nothing but common sense to guide us. New ideas and knowledge, the bloodline of research, are often discovered this way. Such ideas in addition will be practically relevant and significant since they are discovered in the process of solving a real world problem as opposed to extending or generalizing a given mathematical formula. By the same token, when one derives a formula via analysis, it is useful to ask the question: "how is this formula useful to an engineer/user? ". A user is not particularly interested or impressed by the elegance of your derivation but more interested in how the formula will help him; the relative importance of various parameters in the formula; any critical threshold which affects performance, etc. A formula will only be used if the user can understand its significance In commerce, the adage is that to be successful, you must serve your customers well. As an academic in an engineering department or a researcher in an industrial company, we should always remember who are our ultimate customers and make a commitment to serve them well. Finally, a purely theoretical paper no matter how elegant will be thoroughly understood and appreciated by only a few people. But everyone can appreciate the significance of a real world application. For academic engineers, if you want your work widely admired and used, you must adapt your theory to practice and not the other way around. Lastly, most of the published research by most of us belong to the third level which I denote as incremental research within an established framework. These are still important works. These efforts provide continuation and transfer of knowledge from one generation to another, extension, addition and clarification our knowledge base, and above all justification for our existence. To do this kind of research, you must learn the prior literature by consulting reviews, tutorials, and related published papers. However, you should be warned that even if you read the papers on the day they are published, you are at least 18 month out of date due to the delays in reviewing and publishing. This is where your advisor comes in and the reason to go to conferences. An advisor, expert in a given area, is aware of the latest results in the area long before they are published. Other researchers are eager to sent new and unpublished results to these experts/authorities for comments and to establish priority. And to maintain his/her expertise, such advisor keeps up with the literature. If you are students of such as advisor, you benefit accordingly. Secondly, conference proceedings are faster form of publication than archival journals. Also the real benefit of going to a conference is the chance to informally gather the latest information on topics of research. When you are working on the cutting edge of a research topic, you must spend effort to keep up communications with others interested in the same topic. "closing the door in order to built a car (冈门造车)" is not a recommended way to do research which knows no national boundaries. The existence of the Internet ameliorated the problem of information gathering and transmission for the individual to some extent. Students in developing countries and developed countries are more on a even playing field. However, there is nothing to replace the personal attention of an expert advisors and mentor. Speaking about advisors and mentors, there are different philosophies on the role played by a Ph.D thesis advisor. One extreme school of thought (rather prevalent in China but rare in leading universities elsewhere) views the role of an advisor simply as a grader/examiner. Students are supposed to be self-sufficient (自生自灭). The advisor merely makes sure that all rules are observed, requirements satisfied and the student's thesis meet certain minimal standards. This way, a person can supervise ten, twenty, or even fifty Ph.D. students. This makes life easy for the advisor but is bad for the scientific standards of the profession. It also tends to hide the incompetence of the advisor. At the other extreme, the role of an advisor is more akin to that between a kung-fu master and disciple – it is a very serious commitment and responsibility on the part of the advisor lasting over many years and sometimes over a lifetime. In such cases, taking on a Ph.D. student is a time consuming task for at least five years. The advisor works closely with the student and tries to convey and teach many things that are not in the textbooks but important to the student in his/her career and life development. An advisor working in this Socrates' teaching mode can handle at most five or six Ph.D. students simultaneously when he is working full time. My own philosophy is more inclined towards this old fashioned school of thought since I believe except for geniuses, first rate scholars are taught and created this way. As to the standard used in a Ph.D. thesis, I have three: a) Portion of the thesis must be accepted for publication in a leading journal (not just any SCI listed journal) of the field. Top scholars of a field usually agree on what set of journals are leading in a given field. This requirement is not only an independent check of the contribution of the thesis but also serve to publicize the product of the university to the world. b) The advisor should learn something new from the student thesis. c) The advisor should not be ashamed to admit in public that he supervised the thesis since it is also a reflection of the standard and competence of the advisor. Admittedly, two of the three criteria are subjective and can lead to abuse. Only a healthy system of peer review and established self-discipline by the profession can prevent abuse. I also understand the current Chinese regulation for Ph.D. degrees requiring four journal/conference publications. While this is excessive if strictly enforced (footnote: It is more strict than any system in the world and certainly more than my own requirements), I understand to some extent the rationale behind it. Until the quality of Ph.D. advising becomes uniform through out the graduate schools, some form of quantitative requirement will be necessary to prevent abuse and to insure a minimal standard. In short, there is no royal road to research success except focused hard work. |