Monday, December 26, 2011

Coverting Your Outline to a Paper

Once you have an outline for a paper, it is not hard to convert it to a paper. Just start to build up on any section of the outline (other than the abstract and introduction).

Usually, it is easier to start from the system description or methodology. This section contains what you did yourself. So, it should be easier to write. If you have presented some of your work in lab meetings, use those slides as a starting point. If something gets hard to describe, consider using a diagram. If you re reporting lots of boring data, consider using a table.

After you finish writing about what you did, the experiments, and then the results, you can write a rough conclusion based on your results. Having done that, it is time to fill in the related research section. I will cover that in a separate section, for good reasons (you will find out later :-p).

Now have a good break, and read all of what you have written so far. Write a god introduction that prepares people to read all of that. A god introduction is like a funnel. It starts wide, with a sentence that describes something connected to a real-world problem. It ends narrow, with the readers knowing what they are going to read about.

If you have done all these, you can now write a good abstract. Write only WHAT YOU DID. An abstract is not an introduction or a survey. Be clear and brief. Present your best result in a concise manner. One way to do this is to end the abstract with a numerical result (Example: the proposed algorithm recognized 95% of the suspects in the FBI Most Wanted List).

You might have noted that I did not mention the paper length anywhere. If you follow the instructions, you will end up with a paper with a length that is not far from the page limit set by the conference. If there is a difference, it is time to edit the paper. I will cover that in the next section.

Friday, December 9, 2011

Creating an Outline for Your Paper

The best way to start writing a paper is creating an outline. But then, most students think that an outline is some thing like:

Abstract
Introduction
Related Work
System Description
Experiments
Results
Conclusion
Future Work
Acknowledgments
References

This is an outline for MOST papers that have been written and can be written, but it is not an outline for YOUR paper! The reason; it will not get you started.

What you want is a more specific outline, which you can build up with time in to a good paper. Let me explain this with an example.

Suppose you are working on designing ways for disabled persons to use computers. You want allow the user to control the computer by moving their head and face. After a lot of reading, programming, coffee etc, you design a program that can:

1. Recognize if a person is sitting in front of the screen and wants to use the computer
2. Move the mouse cursor on computer when the user moves his/her head
3. Click the mouse button when the user opens and closes his/her mouth.

You also want to evaluate this to see if it is accurate, and easy to use. You design an experiment, and the results show that it is good. So, now you have enough work to write a paper. How do you prepare an outline for this paper?

This is a possible outline that I would have come up with

========================================================
Title: Something like "Evaluation of a Computer Vision-based User Interface for Disabled Users"

Abstract: Write last

Introduction:
Computers are widely used now. But disabled users cannot use keyboard and mouse easily

So it will be good for disabled users if they have alternative interfaces

Computers are getting more powerful, cameras are getting cheaper; so, we can sue computer vision.

Face and neck have lots of muscles, so can specify lots of fine movements. If facial movement can be used to control a computer, more precise control is possible.

In this paper we present a way to control a computer with facial movements. the proposed system recognizes a user wanting to use the computer. it tracks the face, and controls the cursor. We evaluated this system and report results.

2. Related Work
Computer Vision
User interfaces
Vision based interfaces
Nobody has done an interface based on facial movement that is good enough.

3. System description

Overview (need figure)
Face detection (Figure)
Face tracking (equations)
Cursor control (Figure, equations)

4. Evaluation
Experiment design (screenshots etc.)
Procedure

5. Results
Table of typing speed, clicking speed
Discuss why the error is high for the third set of results.

6. Conclusion
Write after the body finished
Add double clicking as future work

Acknowledgment
Thank Roberto for lending the ISO standard documents.

References
========================================================

This took me less than 10 minutes to write, you won't take much longer than that for any outline. Now, all you have to do is fill parts of this outline. I will get to that in my next post.

Note: I actually wrote a paper on a very similar topic, that is why I selcted this example. You can see it here.

Monday, December 5, 2011

Paper Submission Tips - What to Write About

Suppose you have to submit a paper to a conference, or a journal. There can be many reasons for this:

1. You did something new and it works
2. There is this conference that you want to attend, for whatever reason
3. Your supervisor asked you to submit
4. You need publications for graduation, progress review, qualifier, etc.
5. Any other reason

The ideal situation is that the reason is the first. In this case, you can just write what you finished. In case of a conference, you might even write part of what you did so that you can later write a longer article to a journal.

If you want to write because of one or more of the other reasons, things are a bit tricky. To write a research paper, you need to know and/or have done the following:

1. Have a problem to solve
2. Read about whether and how others tried to solve that problem
3. Suggested a method to solve that problem
4. Tried to solve the problem using that method
5. Tested whether your method works
6. Obtained results

If you have done all of the above during the past few months and haven't written about it yet, you are all set. You can start writing.

If you have done only 1 and 2, you can still write a survey paper (preferably to a journal). But for this, you should have read at least 100 papers. This option is for your first year in PhD.

If you have done 1, 2, 3, and 4, you can still write a "short paper," "poster paper," or "work in progress paper" as they would call it. Another option is to start writing while you work on the steps 5 and 6. I don't recommend this, though.

So much for today. My next post will tell you how to make an outline of the paper, using the work that you selected.

Sunday, November 27, 2011

Paper Submission Tips - Part 1

Submitting conference papers or journal articles is one of the most time consuming tasks for grad students. I will post some tips to make this easier for you. Of course, this is going to take more than one post. But each post will be short.

First of all, plan for the deadline. Researchers are not known for meeting deadlines. But it makes our life easier if we do.

If you are a student, your supervisor (advisor) will check your paper before you submit it. Chances are high that he will suggest changes, too. So, plan for the time taken for this. But I suggest that you should not give the prof too much time, either. He/she might come up with additional work to do, and ruin your plans. So, here is my plan.

1. Plan to finish the paper four days before the deadline.
2. Leave it for one day, don't even look at it.
3. Check again the next day, and chances are high that you will notice some mistakes. Correct them.
4. Send to your advisor and co-authors the next day. Two days should be enough for them.

On the day before the deadline, login to the paper submission web site and submit the paper. You can always replace it with the newer version, if your prof has changes to be made.

Most of the time, deadlines are extended. You might see this as a chance to improve your paper, but most of the time it is not. Unless you really, really think so, just leave the paper as it is and start on something new.

So much for now. The next post will tell you how to quickly start writing a research paper.

Friday, November 25, 2011

Qualities of a Successful PhD Student

 Prof. Matt Might of the University of Utah has lost of valuable advice to grad students. This one is short, but a must-read!

http://matt.might.net/articles/successful-phd-students/

You should also read his page more often.

Tuesday, November 15, 2011

Authoring a PhD Thesis: Google Book

Patrick Dunleavy's book on Authoring a PhD is a good resource for those of you who have done most of your work, and are planning to write the thesis soon.

Sunday, November 13, 2011

Where do we get stuck?

If you are doing a PhD, or a research masters, there are a few places where you might have problems. Before getting to details, I will try to list them:
  1. Managing time
  2. Deciding your thesis topic
  3. Interacting with your supervisor/s
  4. Writing
  5. Presentations
A long list, isn't it? Fortunately, things are not that bad. Most students have trouble with only a couple of them. In coming posts, I will try to provide you a few tips on how you can do these things better. Stay tuned!

Tuesday, October 18, 2011

Experiments and Results

 We researchers do experiments and evaluations all the time. Most of the time, these don't give us the results that we wish for. For example, if we design a computer system that can recognize pretty girls by monitoring a camera, it will also recognize several grandmas as pretty girls :-p.

We usually have two choices when this happens. If we are a professor, we can refuse to let our students go on vacation until they improve the results. If we are not, we tinker with the system until the results are more acceptable.

However, there are some occasions where we get really good results. This is supposed to be good, but it actually freaks us out. Most of the time, such good results are a sign of a silly mistake during the experiment, or forgetting to consider something very basic.

A good example of this is the recent news about neutrinos traveling faster than light. Now they have found that they have forgotten that the GPS satellites are moving objects (oh well, it is not like they are rotating around our heads and we see that all the time. Can't blame anyone for forgetting that :-p).

If you are the Einstein-aware type, you can read about the neutrino mystery here:
http://dvice.com/archives/2011/10/speedy-neutrino.php

Thursday, October 13, 2011

Steve, Apple and Research Funding

Computer Science conferences have a list of sponsors, that includes almost all big IT companies. Google, Microsoft, HP, etc, etc. There is one notable exclusion. Apple.

We conducted a thorough survey (okay, asked around) to find out why. Here are the possible reasons:
  1. A survey by Apple revealed that the total financial value of the return from multimedia research in the last five years is 51 cents.
  2. Steve's notebook already has all the "novel" ideas that we presented in the last 15 years, and will be presenting in the coming decade.
  3. Some researchers just wait until Steve comes up with a new device so that they can do research on it. So, they are always behind Apple.
Bye, Steve, we miss you more than all the computer science researchers put together!

Wednesday, July 13, 2011

Best Papers Vs. Rejected Papers

Almost all research conferences give a "Best Paper" award. Let me explain how this is normally done.
  1. If the reviewer thinks that the paper looks very good, he names it as a "best paper candidate."
  2. A few members from the conference organizing committee looks at the best paper candidates, attends the presentations, and select one.
So, the decision is on the hands of about 5 or 6 people. So, let's see how reviewers select a best paper candidate:
  1. If the reviewer is a student (who was asked by the supervisor to do the review), he usually avoids nominating it as a best paper candidate ('not my business'). However, those who do will nominate papers with large numbers of equations that they could not understand at all. Occasionally, they might nominate a paper that the supervisor had good comments about.
  2. If the reviewer is a postdoc, he nominates papers that come from his funding agent, his country, same national as him although from another country, and so on. Most papers are "anonymized" and reviewers are supposed to not know the authors, but practically we know where the papers are coming from.
  3. If the reviewer is a senior professor, he might recommend a  very good paper or a very bad paper. The selection is purely random.
The committee members will usually give an award to a paper coming from a sponsor of the conference. After all, who wants to lose funding or sponsorships during hard economic times?

At ICME 2011, a multimedia conference, we are allowed to vote for the best paper. Nobody counts the number of votes each attendee gives, but the decision is not solely based on votes. Not sure how useful this method will be.

When a paper gets the Best Paper Award:
  1. The authors get a bit of money
  2. They get another sentence on their resume that might allow them to get a better job, get tenured etc.
I am yet to know a best paper that actually resulted in something great. But let me finish by mentioning two rejected papers. They both ended up becoming great things that we all need:
  1. Google's search algorithm
  2. World Wide Web (the Internet, the one that we cannot live without now)