Internal

Contents



Research

Become a good researcher

  • Most importantly: read many papers and excel at math and programming.
  • Keep up with the latest research
    • follow influential researchers/groups
    • read papers at top-tier venues
    • attend talks
    • talk to experts
  • Employ critical thinking
    • what has been done
    • why certain phenomena happen
    • what are the next steps
  • Keep a research log on experimental progress, document successes and failures.
  • Practice quantitative data analysis (statistics) and qualitative data analysis (visualization).

Become a good researcher in a pack (collaboration)

  • Be respectful, responsible, and professional.
  • Understand others' motivations and incentives.
  • Be open-minded to others' topics, thoughts, and approaches.
  • If you do small tasks well, people will offer you big opportunities.
  • Endorse people when you can; being endorsed by an influential person can help you go a long way.

Research process

  • Inspiration → Hypothesis → Experiments → Results → Write-up → Release code/data

Analyze a problem

  • Has this problem been addressed before? (The answer is probably yes to some degree).
  • What are the shortcomings of existing methods?
  • What's your approach? Does it improve the state-of-the-art?

Set up a plan

  • Solving a problem takes time, you need a plan with weekly and monthly goals. Once you have a plan, stick to it.
  • Don't compromise unless some serious limitations are found. Always push the quality of your project to your limit.
  • Always tackle the core problem first, no matter how difficult it is.

Implement an approach

  • Always try to find existing code. If you cannot find any, contact the authors of the approach (paper) telling them how much you admire their work and would like to try an experiment with their code (and data). It's unnecessary to mention your project.
  • Implement an approach from scratch as your last resort: it takes a lot of efforts and it's difficult to ensure the correctness.

Research buddy

  • Meet with a friend regularly to review your progress.
  • Explain your approach to your friend is the best way to organize your thoughts.
  • Discuss the next steps with your friend always results in better solutions.

(Top)



Literature review

The goal of literature review is two-fold. First, it solidifies the novelty and significance of your project. Second, when you write, it shows that you know your area. Both elements are essential for a good paper.

Specific guidelines

  • Review at least 30–50 papers
  • Search keywords in Google Scholar
  • Search keywords in your targeting venue (if you exclude previous work published in your targeting venue, you will be deemed as ignorant).
  • Use directly relevant papers as pivots. Search papers that cite these papers and go through the reference sections of these papers. You can ask experts for pivot papers.
  • For the first-round review, don't read every paper in detail. Read the title and the abstract, if relevant, read the introduction and conclusion, if still interesting, mark it as important.
  • For the second-round review, try to understand the details of important papers found in the first-round review.
  • It's also helpful to keep a reading journal.
    • What is the paper's motivation? Why the problem is interesting?
    • What is the technique?
    • Are there any proofs, can you work out them by yourself?
    • What are the results? How are they evaluated?
    • What are the core messages/key insights of the paper?

(Top)



Publishing venues

The following list consists of our targeting publishing venues along with their h-index and submission deadlines. Machine Learning and Robotics are our main focus.

The h-index is taken from Google Scholar as of July, 2022. Our department considers 40 and above as "high". However, don't take the numbers too seriously. H-index is calculated based on citations, which usually correlate with a venue's size. A small and high-quality conference could have a much lower h-index than a large but mediocre conference.

The submission deadline is just a rough estimate. Please refer to each year's conference website for accurate dates.

  • Machine Learning
  • Robotics
    • RSS (58, Jan)
    • IROS (80, Mar)
    • CoRL (64, Jun), new conference on robotics and machine learning
    • ICRA (116, Sep), flagship conference of IEEE Robotics and Automation Society
  • Control Systems
    • CDC (44, Mar), flagship conference of IEEE Control Systems Society
    • L4DC (N/A, Dec), new conference on control and machine learning
  • Computer Vision
  • Artificial Intelligence
  • Intelligent Transportation Systems
    • IV (59, Feb)
    • ITSC (52, Mar), flagship conference of IEEE Intelligent Transportation Systems Society
    • VTC (Jan and Jul, 42), flagship conference of IEEE Vehicular Technology Society
  • Visual Computing
    • TOG/SIGGRAPH (Asia) (103, Jan and May)
    • VR (37, Nov)
    • CGF (60, Anytime), Pacific Graphics due in Jun
    • TVCG (85, Anytime), Virtual Reality track due in Sep

(Top)



Writing

Writing is as important as solving the research problem: your paper is the only product that the reviewers will be looking at. You need to use the template provided on the conference website and follow the submission instructions EXACTLY. Otherwise, your paper can be returned without review.

General guidelines

  • Follow The Elements of Style.
  • Follow advice here for mathematical writing
  • Outline before you write.
  • Use active voice as much as possible.
  • Prefer words that are specific/definite/concrete to general/vague/abstract.
  • Be succinct, for example:
    • owing to the fact that → since/because
    • in spite of the fact that → though/although
    • call your attention to the fact that → remind/notify you
  • Prefer simple sentences to complex sentences (those with many clauses).
  • Make the paragraph the unit of composition. One paragraph should focus on one message. Start a paragraph either introducing a new topic or helping the transition from the preceding paragraph.
  • Read your writing out loud to check the flow: each sentence/paragraph should flow to the next sentence/paragraph.
  • Write, get edited, repeat.

STARS styling

  • Use present tense as the main tense and use present perfect for the antecedent action.
  • Always use commas before and after "i.e." and "e.g."
  • When listing two+ items, put a comma after the second-to-last item, e.g., A, B, and C.
  • Colon should be preceded by a full sentence. Don't write "I will introduce: A, B, and C."
  • Apply hyphen, en dash, em dash correctly.
  • For numbers one to nine, use words; for numbers 10 and above, use Arabic forms.
  • Write measure in LaTeX as "10~m" or "10\,m", i.e., create an unbreakable space between the number and unit.
  • Use "~" in front of every "\cite" and "\ref" in LaTeX. Don't leave another space between "~" and its preceding word.
    • Correct: "Figure~\ref{}", "Previous study A~\cite{}"
    • Wrong: "Figure ~\ref{}", "Previous study A ~\cite{}"
  • Round every number to one decimal place at the most, e.g., 79.82% → 79.8% or 79%.
  • Use parallel grammatical structure for correlative expressions that are connected by "both, and; not, but; not only, but also; either; or; first, second, third…"
  • Remove "dangling" words of each paragraph.

Technical writing

  • Don't claim something out of your personal opinions; back everything up with evidence.
  • Conduct "pattern matching" throughout your paper.

(Top)



Slides and presentations

You will have various opportunities to present your work. Here are some guidelines and suggestions for preparing your slides and talk.

Treat your talk seriously

  • Your talk may be the only chance for people to put your name, face, and work together. Academia is a small world and you'd want the community to know you in the best way possible.
  • More importantly, think about how many directly relevant papers of your project are out there? When you present at a top-tier conference, guess what, the authors of those papers are likely sitting in the audience. And you want to make a good impression since they are going to review your papers, write your reference letters (even tenure letters), assess your proposals, and refer you to jobs.
  • The goal of your talk is to excite the audience so that they are eager to talk to you and follow up on your work. The best way to impress is to inform. Instead of saying your work is the best, tell people something that they don't know. People will feel their time is worthwhile when they learn something new and important.

Prepare the slides

  • People can only do one thing: listen to you or read your slides. So, your speech and slides should complement each other NOT overlap with each other. This means you should not prepare your slides like lecture notes and, even worse, read them.
  • Everything you put on your slides (text, figures, …) should be visible. Things that can't be seen are useless. The recommended font size is 24+; the minimum font size should be 20.
  • Place page numbers at the top right corner instead of the bottom right corner to reduce the chance of blocking them.
  • Avoid long sentences, use short phrases instead. Incomplete sentences are totally fine.
  • Avoid animated transitions: academia appreciates sharp presentations.
  • Only one slide should be titled "conclusion", use "summary" for other sections at the most.

Content of your slides

  • Motivation (20%), key idea/approach (80%)
  • You must identify a key idea. After the motivation, cut to the chase by talking about the key idea; organize your slides around this idea (prune material that is irrelevant to this idea). It's fine to cover just part of your work.
  • Avoid shallow overviews at all costs. Be absolutely specific about each point and don't leave the audience to figure it out.
  • The best way to explain your idea/approach is to use a simple and intuitive example.
  • People appreciate honesty: make sure to discuss the limitations of your idea/approach.

Speech

  • Watch How To Speak by Prof. Winston
  • Your stance is part of the impression and thus important. Don't move or wobble too much. It gives people a hard time tracking you.
  • You will have two minutes to engage your audience before they fall asleep. Within two minutes, you'd want to address the following items.
    • What is the problem?
    • Why is it significant and interesting? (when the topic involves the suffering of other people, show empathy).
    • Is your approach a worthwhile advance?
  • Show enthusiasm. Enthusiasm is impactful and contagious.
  • If you feel nervous, take a deep breath and move around before your talk.
  • Don't startle in the beginning of your talk, which will make you look unconfident. Script the first few sentences precisely.
  • Don't start with a joke or apologize for not preparing the slides well.
  • Identify nodders and speak to them.
  • Pausing is OK, but don't pause too often.
  • Be specific about the places of your contents when referring to them. Don't just point at a distance. It's difficult to construct that "invisible line" from your fingertip to the content.
  • The audience will "shut down" when your time is up. Continuing is VERY counterproductive. Simply truncate and conclude. Never ask "would you like me to go on".
  • Questions are not an issue but a GOLDEN opportunity for you to connect with the audience. So, pause briefly now and then to encourage questions. If you don't know the answers, just admit it, and ask for pointers afterward.

(Top)



After a paper gets accepted

Congratulations on your paper acceptance! You've done the heavy lifting. Here are the next steps.

Prepare the final (camera-ready) version

  • Address all comments of the reviewers by highlighting (color-coding) the changes. If you have questions, let's discuss. Once you are done, let me know and I'll do a pass of the paper.
  • [optional but highly recommended] Acknowledge people who kindly shared the code and/or data with you, and consider citing their papers as a good gesture.

Watch for deadlines

  • Make sure to submit all required files (final paper, final video, etc.) by their deadlines.
  • Update arXiv if your paper is posted there.
  • Keep track of other deadlines such as the conference registration. We want to register the conference as early as possible, which is usually cheaper.
  • Travel itinerary can be decided later but not too late. Three months prior to the conference is a good time to start booking the travel. See Attending a conference section.

Releasing the code

  • The easier people can access your code, the bigger impact your project will generate.
  • Releasing the code and data also ensures a fair comparison.
  • You don't want an inferior performance of your approach get reported due to insufficient materials for reproducing your experiments.

For more information on how to Release code and data and Add a paper to the website, see Updating the Lab Website.

(Top)



Attending a conference

Travel is exciting. Travel with expenses covered is even better! Here are some items going to be covered for your conference travel:

  • Conference registration and publication fees
  • Airline ticket
  • Hotel (if the destination city is expensive, you may need to share a room)
  • Meals

The hotel and meals will be covered at the most from one day prior to the conference to one day after the conference (department rules). The meals will be reimbursed at predetermined rates (so you don't need to keep the receipts). Other costs such as getting a visa and traveling to nearby attractions are at your own expense. The department understands that our students can't afford to pay for the costs first and get reimbursed later. So, we will try to book many things for you. Meals may be the only thing that requires reimbursement after your travel.

Conference is more about networking than research. Go talk to people and make friends! Conference is also a good opportunity to connect with the industry. Many companies will advertise their openings at a conference.

CS conferences are usually laid-back. You can wear comfortably, although you may want to dress a bit more formally for your talk (no need to go beyond business casual).

(Top)



Fellowships

There exist many fellowship opportunities from various agencies and companies. Winning a fellowship not only grants you higher compensation (usually the case) but also makes you stand out among your peers (it's a great achievement). So, if you spot anything that fits you, definitely apply!

(Top)