Confessions of a researchaholic

2012-06-09

Be the best

Filed under: Real — liyiwei @ 7:16 pm
Tags: ,

This is the most important suggestion I have for research students.

I am only interested in doing the best kinds of work. If I want to publish a research paper, I go for the top venue (e.g. SIGGRAPH for graphics or CHI/UIST for human computer interaction). There is nothing inherently wrong about not aiming for the best; it is just not what I want to spend my time on. Life is too short.

I do not care where you came from or what kinds of performance you had before; if I take you as my student I am going to help and push you to be the best. I will not allow any excuses. Do not join me if you are not fully committed to this.

Some frequently asked questions below.

Q: why?

I am actually not sure how to answer this question, because it is an integral part of who I am. (My first name literally means “the best” in Chinese.)

But if you insist, here are some reasons that I can pull out of my head.

I just see no fun and no point otherwise. Imagine asking an ugly girl/guy out and getting rejected. Go for the prom queen/king.

This is the best kind of training, both mentally and intellectually. If you can pull off the best kinds of research work, you will gain the confidence to tackle whatever other challenges you will face in your life.

This is a transformative kind of experience. There are many very smart and hardworking folks out there. To compete with them, I have to be at my best. And this requires me to eat well, sleep well, exercise a lot, maintain good personal relationship, etc.

I want my students to land the best kinds of jobs after graduation. And the quality of jobs is positively correlated to the quality of publications.

Q: how about people who do not have enough talents?

Talent is overrated, especially for endurance sports like research. What matters most is passion and toughness. I have seen more people who lack hearts than who lack brains.

None of my (current and past) students attend Stanford, but many of them can perform in Stanford level. I just need to bring out the best of each one of them.

Q: what if I fail?

Failure is a natural part of research; if you do not like the former, you should not undertake the latter. However, even in the very worst cast that you have all your papers rejected, if you can develop solid programming skills, you should have no problem finding a good programmer/engineer job.

Q: how do I know what it takes?

Talk to my current and past students, and if possible, try to work with me for a little while before full commitment.

2012-06-01

Resume blogging

Filed under: Real — liyiwei @ 10:31 am
Tags: ,

After giving a recent talk titled “how to have fun in your research and get away with what you want in life”, I started to get a lot of questions, on top of all these that I already have from the ordinary student recruiting/application process.

To maintain my sanity, I am going to summarize my advising style and answer all the common questions, (hopefully) once for all.
Stay tuned.

2012-03-29

Details

Filed under: Imaginary,Real — liyiwei @ 7:33 pm
Tags: , ,

God is in the details
Devil is in the details
So which one will be there
When we look into the details

2012-03-26

Another data point

Filed under: Real — liyiwei @ 1:49 am
Tags: ,

In responding to my earlier post, a very talented graphics researcher has shared with me his statistics, as shown below.

As you can see, his has much better success rate than me, especially considering that he has been doing rendering, a tough field for SIGGRAPH. 🙂

2012-03-24

Fooled by randomness

Filed under: Real — liyiwei @ 11:41 pm
Tags: , , ,

Feeling down from some recent rejections? I hope this post will make you more positive. The gist: never, ever, quit.

Assume you are throwing N loaded dices, each with a probability p for coming up head.

Now, if p is greater than 0 and smaller than 1, there is always a chance that the N dices will come up with all heads or all tails. And the smaller the N value, the more likely for such extreme cases to happen.

This is all pure chance. But unfortunately, human brains have difficulty accepting randomness, and always want to impose determinism, e.g. patterns or rules or causalities.
For example, if you are a scholar submitting N papers to a conference, you will likely consider yourself to be very good/bad (or the paper committee has treated you very well/badly) if all N submissions are accepted/rejected.

This human fallacy is brilliantly illustrated by Nassim Nicholas Taleb’s book “Fooled by Randomness”.

However, even without reading that book, I can recommend a very simple remedy: law of large numbers. This is a well-known mathematical theorem, which says that the expected value of a random variable can be more accurately predicted by averaging a larger number of samples.

So, for example, to measure your intrinsic paper acceptance rate towards a specific conference, you can take the total number of acceptances divided by the total number of submissions. This will be a much more meaningful measure than your acceptance rate for a single year, especially if you have a sufficient number of submissions across multiple years.

For example, the plot below shows my cumulative acceptance rate for SIGGRAPH, the top venue for computer graphics and interactive techniques. As you can see, the rate seems to be gradually converging to a certain value, around 0.34. This is much more stable measure than my yearly rate, which can be anywhere between 0 and 1.

Now, if you are new to a field, your rate will have a higher variance, just like the initial portions of mine. I was lucky that I had a good start which boosted my confidence. (Initial condition is actually very important and has been found to greatly influence the performance of many careers, e.g. hockey players. Note to myself: dig out that book/article. I guess it should be Geoff Colvin’s “Talent is Overrated” or Malcolm Gladwell’s “Outliers”.) However, if you happen have an unlucky start, do not give up too early; hang on for a while, so that you can have a chance to see your *intrinsic performance*.
As you can see, my intrinsic performance did not really show up until about a decade doing SIGGRAPH.

(With all these rational arguments, I have to confess that it still hurts to get rejected!)

Some notes about the graph: (1) I plot SIGGRAPH at integer years and SIGGRAPH Asia at integer + 0.5 years, (2) missing data points are for years which I did not submit anything (2004 and 2005 while in NVIDIA and 2011.5 when I have nothing to submit for SIGGRAPH Asia 2011), (3) a more accurate measure would be “moving average” (with exponential decay of past values) but I probably need another 10 years to warrant this, (4) I really want to improve my intrinsic rate to at least 50 percent!, (5) I guess the ultimate test is to have multiple disjoint committees + reviewers, all with similar qualities, to evaluate the same batch of submissions, and see if they will accept similar sets of papers.

2012-02-22

Artificial intelligence

Filed under: Imaginary,Real — liyiwei @ 10:37 pm
Tags: , , , , , ,

When I was younger I preferred to stay away from people as much as possible, as most of them are not very interesting and it is much more rewarding for me to be alone thinking and reading.

When I get older, I realized that humans are intensively intriguing subjects for study. I started to spend a lot of time observing human behaviors and try guessing what they are thinking and predicting their actions.

This caused certain dilemma for me: on one hand I still want to be as far away as possible from people, but on the other hand, I want to be close enough with them for the purpose of studies and observations.
(The penalty and reward seem to go in tandem; crowd behavior is the most interesting, but also the most annoying to be part of.)

Fortunately, computer science comes into rescue. Far from the common stereotypes (of nerds locking in toilets), computer science, especially the most current and active subjects, are very human centric. One example is user interface, including design for better user experiences, as well as analysis and synthesis for deeper understanding and more advanced interactions.

A more recent example is social networking. Previously, most human daily activities simply dissipated into entropy. Now, with people spending more of their interactions through various social networking sites, we can record their activities in better quality and quantity.
Such data not only enables better computer technologies but more profoundly, more insights into human nature. (Facebook probably knows more about certain individuals than their mothers do.)

Two sci-fi series could provide inspirations for both directions.


Caprica is about how humans create Cylons, a cyber-genetic life form that eventually pushes humans near extinction in the main Battlestar Galactica series (which I found to be much less interesting).


Dollhouse is about how technologies can allow memories and personalities to be extracted from one individual and installed into another, essentially programming human brains.

Both offer insights into computer science and humanity, as well as highly enjoyable entertainments. Unfortunately, both got canceled prematurely due to low ratings, a confirmation of my childhood observation about how ordinary humans would react to deeper materials.

2012-01-17

Representative image for SIGGRAPH submission

Filed under: Real — liyiwei @ 5:24 am
Tags: ,

Basically, its main function is to be displayed on the front of the room during the committee meeting, so that people can have a visual reference/cue of the paper being talked about.
Even though the representative image does not have any practical value for the paper (it is not illegal to use an arbitrary image, like, say, picture of Lindsay Lohan, at least with clothes on), the committee members are humans and thus can be subtly influenced by all psychological factors.
Thus, it is to your advantage to use an image that can help your paper, i.e. conveys the main ideas or results while looks appealing.

2011-12-30

James Landay on “China Will Overtake the US in Computing…Maybe, Someday…”

Filed under: Real — liyiwei @ 7:47 pm
Tags: , , , ,

This is a mandatory reading for all my (current and future) students with an initial Asian training:

James Landay on China Will Overtake the US in Computing…Maybe, Someday…

First of all, let me share one of the biggest secrets of China (and to some degree other Asian countries like Japan and Korea as well as ethnic Chinese states like Taiwan and Singapore). Do you ever wonder why China developed this authoritarian culture in the first place? It is very simple: a conforming population is much easier to rule than one that can think freely. The Chinese emperors were very calculating on this; they did not even allow alternative sources of authorities to challenge them (like the bishops who can thorn up to European emperors’ arses). On the other hand, they also want the population to be reasonably fluent so that the country can be productive. Thus the duality of the Chinese/Asian education system: on one hand it enforces conformity, and on the other encourages intellect and hard working.
Unfortunately, even though this system worked for the past agriculture and manufacture dominant economic systems, a knowledge-based economy will require citizens who can think. So China will have to change its culture and education systems, or face competitive disadvantages.

Part of the fun for my past MSR and current HKU posts is being close enough to help while far enough to not get dragged down into the sinkhole. I am curious how much I can do as an individual, or there is really some grander scale environmental stuff that I simply cannot reproduce. Results so far are very encouraging; Asian students who worked with me for sufficiently long periods of time (at least one SIGGRAPH cycle) have shown significant progress of thinking skills and at least one of them managed to create SIGGRAPH ideas.

For the sake of more fun, I now extend my grand challenge to MSR Asia to all my (past, current, and future) students: the first one to publish a single-authored SIGGRAPH paper will receive my full financial support, out of my own pocket instead of any grant, to make the trip. (Really, it is not that hard; I am not very smart, and I did that twice already. I make this challenge because only with a single-authored SIGGRAPH paper can you prove your full independence, including creativity.)

2011-12-28

Writing a previous work section

Filed under: Real — liyiwei @ 2:23 am
Tags: ,

Every research paper has to cite and describe relevant previous works, often in a dedicated previous work section, or sometimes as part of the introduction.
Trying to write a previous work section can be an excellent mirror exercise for reading papers. The quality of the former can be a good indication of how much you really understand from the latter.

First of all, spend enough time reading how other people write previous work sections. Pay particular attention to those you find educational, from which you can actually learn something, as well as those you find unhelpful, from which you feel only confusion and frustration.

A good previous work section contains multiple layers of information with different resolutions of details.

  • At the bottom level (highest resolution), you describe in one or a few sentences about the gist and merits of each paper.

  • At the middle level (medium resolution), you organize related papers into paragraphs, each contains information about the gist and merits of the group.

  • At the top level (lowest resolution), you relate every citation with your paper, in particular, how yours stands on the shoulders of previous giants: commonalities, inheritances, and mostly importantly, innovations.

After you have a draft, show it to your adviser or other experienced researchers, ask for advice, and iteratively improve it. It will be an excellent exercise for both reading and writing papers.

« Previous Page — Next Page »

Theme: Rubric. Get a free blog at WordPress.com