Statement of Purpose

randombroadAI and Robotics

Oct 15, 2013 (3 years and 9 months ago)


Statement of Purpose

In the next five to ten years, I see myself as an operations researcher, giving significant input
to help solve the challenging problems facing business, finance and manufacturing industry. It is my
nature to solve problems, figure out

how things work, and figure out how to make them work better.
Last year, my focus was attracted by the power of machine learning and yet as I discovered further
the reciprocal relationship between machine learning and operations research, I was more
nced of pursuing a master degree in the operations research program.

I am passionate with this field because operations research interweaves tightly with my
academic growth in college. The past three years witnesses my attendance via mathematical

contests and self
motivated experiment that follows. In this involvement, I get the chance
to work with many interesting people; in this involvement, I find my truly supportive friends; and in
this involvement, I experienced the joy of exploration without

an answer beforehand, of struggling
against deadline and of my sense of achievement with or without recognition. All of these
possibilities would have otherwise remained invisible to me and now I take it an integral part of my

As my CV reveals, my
team's research covers various disciplines from engineering to logistic
to biology. As a hallmark, our self
motivated study in modeling was appreciated by a professor who
granted us a special ticket as one of thirteen graduate
level representative teams fo
r our university.
In addition to that, it is also worth mentioning of our first giant step, which is the development of a
rescue scheme. After our exploration of the full complexity with up to seven models, I realized the
possibility of further investigati
on with one of our central model, which is, finding a Hamilton path
with least turning delays. To give a more elegant solution than our previous heuristic one, I learned
to consult related ACM/SIAM literature, formulated the problem and assembled more tale
nts. We
worked for weeks to

finish a paper that provides a

refined algorithm for finding a Hamilton path if
constrained in rectangular grid graphs. These series of work attested to myself a potential to do real
serious research.

At that point, I knew I we
re prepared to continue the career in operations research with the
obtained knowledge, yet what further reinforced this decision is actually a desire to lay a foundation
for my further involvement in an interdisciplinary realm between operations research a
nd machine

I think we often get excited at those things we see their value and meantime see they are
within reach. And that is what data mining is to me. Last term I participated in the Knowledge
Discovery in Databases Cup (KDD Cup) 2011, an int
ernational data mining contest held on
The task of our choice is to discriminate between music tracks (provided by Yahoo! Lab) highly
rated by certain users from those which are overall highly rated. There was a dramatic change in my
role. While othe
r members claimed at first to have known data mining softwares such as SAS, SPSS
and Clementine, it turned out that they had to rely on me to do the data manipulating with
MATLAB and C so that we could use. Furthermore, I became the core member as I learne
d to
assume those ready
made softwares' implementations exclusively in MATLAB with super fast
efficiency. I programmed to perform clustering, regression, interaction and visualization with BLAS
architecture in mind. I picked up strtok() command from lines
and lines of programs on
line to
process raw data amounted to 1 GB. And more encouragingly my analysis and comparison proved
to be helpful as I chose the right cluster parameters and realized the potential value of scoring time,
as the winner (National Tai
wan University) did.

I had expected that people could do a good prediction with so much data in hand, yet I hadn't
expected they could do as accurate as with 2% error rate! Throughout the conference proceedings, I
found matrix factorization and optimizati
on concepts all over the place. You know, that's what I had
learned from courses and books that I had no idea how to use them before. Furthermore, in a
colloquium given by Prof. Ya
xiang Yuan (Chair of Operations Research Society in China) I learned
more t
hat the problem had a similar background with the famous Netflix problem, which had a
concise mathematical representation
matrix completion with lowest rank. In discovering all of this,
I know that data mining/machine learning will afford me the joy to us
e my analytical skills,
research experience and creativity to a field full of imagination. To that end, I need to study further
about optimization, modeling and engineering (e.g. signal) in a operations research program.

You might wanna ask "Judging by yo
ur research experience, why not a PhD program, where
you become famous while paying zero tuition?" My answer is I do better job to connect cases in
their conceptual layer than the theoretical layer. In other word, I would like to be an expert who
s and transplants success as opposed to who creates archetypes out of pure theory. No
concreteness, no conception; No experience, no motivation. That's my way, I think, and my recent
industry experience reinforces this notion.

(Why Stanford?) I consider S
tanford my dream school. Beyond fame, I see openness,
meticulousness and erudition. Currently I am attending the Stanford's Machine Learning open
course (Advanced Track) instructed by Pro. Andrew Ng. I get so impressed by his way of teaching,
the abundance

of materials and good feeling when I can share insights and knowledge with millions
of talents across the world. You know, I can't help doing those pedagogical exercises until late at
night. I strongly believe that the strength of a combination of operati
ons research and machine
learning in Stanford will arm me to conduct interdisciplinary research in my future career.

(Why Berkeley?) I love to pursue my graduate study in Berkeley mainly for the strength of
the Department of IEOR in this field. In fact, I

am a kind of surprised to find the membership
Papadimitriou who authored the major reference of our mentioned paper, which attests to the
caliber of the program. In addition to that, Berkeley is also famous for her renowned computer
science and mac
hine learning lab. I strongly believe that the strength of a combination of operations
research and machine learning in Berkeley will arm me to conduct interdisciplinary research in my
future career. Besides, the climate in California is really pleasant.

ooking back at it, I have prepared myself with solid mathematics background, great
familiarization to cross
disciplinary research and a strong will to learn more. I once studied in
UCLA summer session in California. At that time, I didn't get the chance to

visit Stanford. This
time, I hope I will.