How to be a Winner
Advice for students starting into research work
View this page in
courtesy of Nick Stasov,
courtesy of Valeria Aleksandrova,
courtesy of John Vorohovsky,
Swedish courtesy of
courtesy of Jordan Silaen,
courtesy of Philip Egger, Vietnamese
courtesy of Laura Himmer.
[N.B.: Observations and recommendations based on first being a
UROP student at MIT and later supervising numerous UROP students at
MIT and undergraduates and graduate students at UCB.]
Don't get hung up trying to understand everything at the outset
The biggest challenge you face at the onset of any new project is
that there is a huge (seemingly overwhelming) amount of stuff you need
to know to tackle your problem properly. While this phenomenon is
true in the small for the beginning researcher, it is also true in the
large for any research project. So learning how to cope with this
challenge is an important skill to master to become a good
researcher. In contrast, blocking your action and progress while
waiting for complete knowledge is the road to failure.
Coping mechanisms employed by winners include:
Losers will stop the first time they run into something they
don't know, cannot solve a problem, or encounter trouble
slightly out of what they consider ``their part'' of the problem and then
offer excuses for why they cannot make any progress.
- prioritizing (what do I need to know most)
- read (everything made available to you, and seek out more; but
don't put months of reading between you and getting started
- multithreading (when blocked on one item or path, is there
another I can productively pursue?)
- pursuing multiple, possible solution techniques (maybe
some have easier/less blocks paths than others)
- wishful thinking (ok, let's assume this subproblem is solved,
does that allow me to go on and solve other problems?)
- pester people who might have some of the information you need
(you might think they should know what you need to know,
but often they don't have a clear idea of what you do and don't know;
start by getting them to give you pointers to things you can use to
help yourself. Show respect for their time and always follow up
on the resources you've been given before asking for a personal
- propose working models --- maybe they are wrong or different
from others, but they give you something to work with and something
concrete to discuss and compare with others. You will refine your
models continually, but it's good to have something concrete in mind
to work with.
Winners consider the whole problem theirs and look for
paths around every hangup.
Losers make sure there is someone or something to blame for
their lack of progress.
Winners find ways to make progress despite complications.
Losers know all the reasons it cannot be done
Winners find a way to do it.
Communicate and Synchronize Often
Of course, when you do have to build your own models, solve unexpected
problems, make assumptions, etc. do make sure to communicate and
synchronize with your fellow researchers. Do they have different
models from yours? What can you learn from each others' models and
assumptions? Let them know what you're thinking, where you're stuck,
and how you're trying to get around your problems.
The whole problem often seems overwhelming. Decompose it into
manageable pieces (preferably, with each piece a stable intermediate).
Tackle the pieces one at a time. Divide and conquer.
This may sound obvious, but it works. I've turned numerous problems
which appeared ``frightening'' in scope into many 1-day or 2-day
tasks, and then tackled each nice, contained 1--2 day task as I came to
it. As I understood more, new problems and tasks arose, but they
could all be broken back down to bite sized pieces which would be
tackled one at a time.
In computer systems especially, the biggest limitation to our ability
to conquer problems is complexity. You need to work continually to
structure the problem and your understanding of it to tackle the
inherent complexity. Keep careful track of what you have done and
what you need to do. Make lists; write it down; don't rely on your
memory (or worse, yet, your supervisor's memory) to hold all the
things you need to do and all the intermediate issues you need to
Make priorities in your efforts and check your priorities with your
supervisor. A common occurrence is for your supervisor to ask you to
do A, forget about it, and then ask you to do B before you could
possibly have finished A. If you are uncertain on whether B should
take priority over A, definitely ask. Sometimes it will, but often it
won't, and your supervisor will be glad that you reminded him you were busy
solving A. Keep track of B, and when you finish A, see if B still
makes sense to pursue.
Realize that your supervisor is busy
Your professor or graduate student supervisor is busy. He hired
you to help him get more accomplished than he could have on
his own. Your biggest benefit to him is when you can be self moving
Do not expect your supervisor to solve all your problems.
Find out what he has thought about and suggests for a starting point
and work from there. But, realize there may become a time when
you have put more quality thought into something than he has (and this
will happen more and more often to you as you get into your work).
So, when you think you see or know a better way to solve a problem,
bring it up. In an ideal scenario this is exactly what should
happen. Your supervisor gives you the seed and some directions, then
goes off to think about other problems. You put in concentrated time
on your problem and ultimately come back with more knowledge and
insight into your subproblem than your supervisor.
As a supervisor, I work in two modes:
- Until a student has demonstrated that he has thought more deeply
about the problem than I have, I strongly advocate that he start things my
- Once a student has examined a problem in depth, then we
can discuss it as peers, and generally the student becomes the
expert on this subproblem, and I can offer general advise
from my experience and breadth.
Once you've signed up you have to deliver. But, you do not have to
deliver the final solution to everything at once. This, in fact, is a
fallacy of many people and research projects.
Losers keep promising a great thing in the future but have
nothing to show now.
Winners can show workable/usable results along the way to the
solution. These pieces can include:
Demonstrate progress. This allows your supervisor to
offer early feedback and to help you prioritize your attention---this will
often help you both make mid-course corrections increasing the
likelihood you will end up with interesting results in the end.
Requirements and understanding invariable evolve (remember the
key challenge at the beginning is incompletely knowledge). Change and
redirection is normal, expected, and healthy (since it is usually a
result of greater knowledge and understanding). The incremental model
is robust and prepared for this adaptation while the monolithic (all-at-once)
model is brittle and often leads to great solutions which don't address
the real problem.
- solutions to simplified models
- pieces of a flow
- intermediate output/data
- measurements of problem characteristics
- stable intermediates (see below)
Incrementally grow your solutions (especially software). In
the new chapters which appear in the 20th Anniversary edition of
Mythical Man Month, Brooks identifies incremental development
and progressive refinement towards the goal as one of the best, new
techniques which he's come to appreciate since the original writing
of MMM. From my own experience, I
whole heartedly agree with this, and it does have a very positive
impact on morale (yours, your team's, your supervisor's).
Target stable intermediates
Look for stable intermediate points on your incremental path to
solving some problem.
- points where some clear piece of the problem has been solved
(has a nice interface to this subproblem, produces results
at this stage)
- things you can build upon
- things you can spin-out
- things you can share with team members (allow them to help)
- points of accomplishment
Don't turn problems (subtasks) into research problems
Often you'll run into a subtask with no single, obviously right
solution. If solving this piece right is key to the overall goals,
maybe it will be necessary to devote time to studying and solving this
subproblem better than it has ever been solved before. However, for
most sub-problems, this is not the case. You want to keep focused on
the overall goals of the project and come up with an ``adequate''
solution for this problem. In general, try to do the obvious or
simple thing which can be done expediently. Make notes on the
possible weaknesses and the alternatives you could explore
should these weakness prove limiting. Then, if this does become a
bottleneck or weak link in the solution chain, you can revisit it and
your alternatives and invest more effort exploring them.
Learn to solve your own problems
In general, in life, there won't always be someone to turn to who has
all the answers. It is vitally important that you learn how to
tackle all the kinds of problems you may encounter. Use your
supervisors as a crutch or scaffolding only to get yourself started.
Watch them and learn not just the answers they help you find, but how
they find the answers you were unable to obtain on your own. Strive
for independence. Learn techniques and gain confidence in your own
ability to solve problems now.