490 likes | 638 Views
Byron, Dan, Zan, and Jennifer Acknowledgement: Hundreds of meetings with Dr. Chen. A Template for Producing Research Papers in the AI Lab. Disclaimer: Use this advice at your own risk. If we already knew how to do it, it wouldn’t be research, and we would already have tenure!
E N D
Byron, Dan, Zan, and Jennifer Acknowledgement: Hundreds of meetings with Dr. Chen A Template for Producing Research Papers in the AI Lab Disclaimer: Use this advice at your own risk. If we already knew how to do it, it wouldn’t be research, and we would already have tenure! Disclaimer 2: This presentation is filled with student perceptions of what Dr. Chen said or meant, and spiced with our own commentary. These views may or may not accurately reflect Dr. Chen’s position when you turn in a paper or give a presentation.
Notes on the Title Abstract Introduction Background and Literature Review Research Questions Research Testbed, System Design Research Design Experimental Results Discussion Conclusions and Future Directions References Some additional suggestions Reviewing papers Agenda
Notes on the Title • A good title should reflect the entire contents of the paper, avoid using cute titles. • Should be less than 8-9 words. • Complete sentences are not necessary, use colons if needed. • Acronyms • If needed, select meaningful acronyms that convey the meaning of the work/project (e.g., COPLINK, GeneScence). • Acronyms help in branding a system/project. • Do not forget acknowledging the funding agencies on title slides of presentations!
One of the most important parts of the paper. Concisely express the problem in one sentence or two. Mention why the work is important (if the goal of the research was accomplished, what good thing would happen… ) Describe methodology Highlight the most important results Gives the reviewer a reason to continue reading the paper Should consist of short sentences Don’t invent new words! * Slides do not generally include an abstract * Abstract – What is in it?
Publish, Publish, Publish (or Perish!) Although many different methodologies can lead to an accepted publication, these principles should be particularly useful to you in the AI Lab environment. We will present our view of how to get papers past Dr. Chen and out for publication. The content is presented in the form of a template. Abstract – Our Abstract
The Introduction “hooks” the reader. What is the motivation for the work? What is the context? Big picture, how did we address the problem? What will the rest of the paper look like? Introduction should be 3-5 paragraphs A four page introduction is not a good introduction The last paragraph always points to the structure of the paper Introduction – What is in it?
PhD students want to learn to be productive researchers. A number of research approaches are effectively employed by various researchers and research groups. The model used here in the AI Lab has a strong track record of producing work publishable in top tier journals. Introduction – Motivation/Context
To be productive we need to be: Doing the right research, Doing the research right, and Packaging the research appropriately for distribution. Missing any of these elements substantially reduces the chance that your work will be published. Introduction – The Big Picture
Criteria for choosing a research project that might eventually be funded. Does it advance Science? Choose work that makes a contribution to some scientific body of knowledge. Implementing an effective system is not enough: the methodology and techniques are important. Does it have potential impact? Do work that will improve important real-world processes. Thus, we emphasize domain-specific applications and completing a line of work in a user study. Caveat: Making an impact with one paper is tough. Doing the Right Research 80% 20%
Research should be publishable. Strong methodology is vital. Are the experiments rigorous and valid? Precise hypotheses Ideally, hypotheses are based on previous literature or established theories Appropriate statistical tests Even if the contribution is small, good methodology can get a paper over the top. Methodological flaws give reviewers an excuse to reject your work. Doing the Research Right
The should be publishable, not necessarily suited for funding. Extend your current work, don’t go down an entirely different path. Choose wisely in your Ph.D. Use your methodology on other collections and different contexts. Define an area, so people know you for your work. Right Research for Junior Faculty
Even solid work will be rejected if it is not appropriately packaged. Main ideas of good packaging: Be concise Be professional Target the journal or conference Be persuasive Appropriate Packaging
Good Slides Force you to organize concisely and clearly, Allow Dr. Chen to present the work to keep the $$$ coming, and Reduce the time needed to write a good paper. Good writing Describes previous work in a digested form Does not distract the reader Makes a coherent argument Employs good examples to illustrate difficult techniques or concepts The transition from good slides to a good paper is 2-3 weeks. Appropriate Packaging
In future sections we will: Review the main points: Lit Review Present our Research Questions Describe important environmental issues: Research Testbed Discuss methodology: Research Design Present our Results Discuss the implications Draw Conclusions and look to Future Work Introduction – Looking Ahead
Abstract Introduction Background and Literature Review Research Questions Research Testbed, System Design Research Design Research Findings Discussion Conclusions and Future Directions Additional Suggestions Reviewing Papers Agenda
One of the most important parts of a paper The lit review frames the work. Connection between introduction and research questions Introduction points out the motivation Literature review provides more evidence of the limitations in previous studies Following this logic flow, lit review leads to specific research questions What to do? e.g., new algorithm, performance, etc. Literature Review: What is it? Although, a paper is sometimes accepted largely because of a strong lit review that summarizes and organizes an area of inquiry
The literature review presents “digested material” Taxonomies/Frameworks are good A taxonomy of 2~4 dimensions Know all studies in the field and focus on relevant ones Tell what previous work means Choose the right/relevant subset of all the papers you could cite Don’t try to review everything, understand the audience of the paper Literature Review: Key Idea
Different level of completeness depending on journal/audience In general, a more comprehensive review For special issues, not too big, more focused Different focus depending on your research question To propose a new task, To compare performance, or Etc. Literature Review: Completeness
NOT*: “use too much tutorial,” “educating the reader,” “lose the seminal works,” “making sure we mentioned everything” (No laundry lists!), or “too critical to others’ work” (Maybe he/she is the reviewer) INSTEAD: Enough coverage Be Comprehensive Critiques (what are missing leads to your research questions) Show why our approach makes sense Provide a benchmark for comparing our results Literature Review: Tips * Although, a paper is sometimes accepted largely because of a strong lit review that summarizes and organizes an area of inquiry
Abstract Introduction Background and Literature Review Research Questions Research Testbed, System Design Research Design Research Findings Discussion Conclusions and Future Directions Additional Suggestions Reviewing Papers Agenda
The intro said why. The lit review set up the argument. Research Questions: focus the work suggesting what we can measure, follow logically from the lit review (address the critics in lit review and lead to your findings), and should be answered by the experiment(s). Research Questions
Have 2-5 major research questions, They should have clear scientific motivations: Innovation to basic science, and potential impacts. Research Questions
Abstract Introduction Background and Literature Review Research Questions Research Testbed, System Design Research Design Research Findings Discussion Conclusions and Future Directions Additional Suggestions Reviewing Papers Agenda
What data sets will be used in the experiment(s)? Testbed should be interesting, relevant, and significant. We have: Slides available from previous presentations Published papers Research Testbed
Describe how the architecture works and its components The basic publication flow: System Design Write Grant Proposals Topic Identification Format and Submit Prepare Initial Slides Revise and Write Revision Letters Experimentation Final Slides Presentations Conferences & Funding Agencies Write a Paper
Good diagrams help readers understand better and clarify our own thoughts. Behavioral papers may have a methodology section instead of system design. Algorithm papers discuss methodology and algorithm design (pseudo codes and diagrams are suggested) in this section. Methodology needs to have a theoretical foundation. System Design
Abstract Introduction Background and Literature Review Research Questions Research Testbed, System Design Research Design Research Findings Discussion Conclusions and Future Directions Additional Suggestions Reviewing Papers Agenda
Focus on the experiment (s). Present hypotheses Measurable Address the research questions Plan for statistical tests Research Design: What’s in it?
Is to validate your research. Use credible experiments to verify the hypotheses. Methodology: Quantitative measures: such as accuracy and speed. Qualitative measures: explains the inside phenomena of the quantitative results. Simulation: is often used in system design arena. Research Design: What’s in it?
Abstract Introduction Background and Literature Review Research Questions Research Testbed, System Design Research Design Research Findings Discussion Conclusions and Future Directions Additional Suggestions Reviewing Papers Agenda
Tables and figures are critical. Need to be consistent and neat. Highlight interesting numbers. In caption, you may use 3-4 sentences to describe more details about a figure or a table. Use a small paragraph in text to explain the essence about a figure or a table. You may group your findings in chunks, each of which starts with a bold summarizing sentence. Research Findings
Abstract Introduction Background and Literature Review Research Questions Research Testbed, System Design Research Design Research Findings Discussion Conclusions and Future Directions Additional Suggestions Reviewing Papers Agenda
The discussion section gives meaning to the results. Why did you get the results you got? If some of the results were surprising, why? What did you observe outside the measured information presented in the research findings section? Discussion
Can have some duplication with the abstract. State the contribution, but don’t overstate it. Don’t form questions in the reviewers mind. Don’t mention trivial future directions. Point to several promising directions. Conclusions and Future Directions
Where has similar work been published? What kind of articles are accepted by the target journal? Remember who did previous work. Know where it was published. Try to reference related papers that were published in the target journal. Must have 5-10 key journals, key conferences in the field Number the references Have a consistent format References
Abstract Introduction Background and Literature Review Research Questions Research Testbed, System Design Research Design Research Findings Discussion Conclusions and Future Directions Additional Suggestions Reviewing Papers Agenda
Eliminate typos and grammar errors Consistent formatting Clear figures and tables Captions make the meaning of the figure clear The layout should be clear and clean Every figure/table must be referenced in the text Professionalism
Control your time; 40 slides not 75! Present with energy and enthusiasm Listen to questions; you can clarify before you answer Don’t avoid questions, especially if they ask for specific information. Rehearse; know what slides come next Don’t read from your slides Presentations
Be concise: Do you need this slide/sentence/word/paragraph? Is the prose wordy? Active sentences are better than passive. Flow: Why is this point here? Present info in a logical, top-down flow Good: Therefore….. Bad: You’ll see why later…. Be concise. Consider the flow
What is the problem? Why do we care? How has it been addressed before? What is the research gap? How are you going to address the research gap? It should be clear from previous material or input here why you chose each part of your solution. An example of good flow (1 of 2)
How will you measure the results? What were your results? Were they statistically significant? What did you learn? Why is that important? What will you do next? An example of good flow (2 of 2)
If you vary from the template you should have a good reason. Dr. Chen resists incomplete slides. Multiple revisions improve your slides. Get through a couple of revisions before you show them to Dr. Chen. Let your colleagues help. More things to think about
Be ready with a good set of slides 3 days ahead. Practice presenting them. Tighten and revise. Have a final version no less than 24 hours in advance. Practice presenting the slides! A challenge:
Abstract Introduction Background and Literature Review Research Questions Research Testbed, System Design Research Design Research Findings Discussion Conclusions and Future Directions Additional Suggestions Reviewing Papers Agenda
Be professional Pretend everyone will see your review Be accurate Be specific Be critical Of the methodology Reviewing Papers
Read similar papers Summarize the paper Separate major and minor comments Your review becomes your reputation Don’t rewrite the article Reviewing Papers
1-3 pages (never less than ½) Include high level… Summary Strengths Weaknesses Selection Categories: Accept as is NEVER choose this one Indicates laziness (Yours!) Minor revisions No methodology problems Findings are interesting Journal Reviews
Journal Reviews • Selection categories cont… • Major revisions • Paper can be fixed through a new experiment • Needs significant clarification • Is incomplete • Rejection • Wrong methodology/implementation • Findings are trivial/uninteresting • Have I learned something new? • Wrong Journal • Suggest a different journal • Not substantial enough, recommend as a short note • Complete Journal Reviews in 2-3 months
Conference Review • ½-1 page • Accept only if light editing is necessary • No time for major overhauls • Is research and methodology interesting? • Complete conference reviews in 2-4 weeks