Related Works Search

Requirements

Create a new web-page with the following information.

  • At least five related works (research, blog posts, videos, books, etc.).
  • At least two should be research articles.
  • For each work, you should provide a one to three sentence summary.
  • Add a few sentences about how your project is different than these related works.

With the related works, try to answer the question: what is the state-of-the-art in the domain of your project?

This is similar to an annotated bibliography, but I am not expecting you to write more than a few sentences for each entry.

Final reports will require more than two research articles, and you should plan on collecting/skimming additional related works throughout the semester.

You will submit a link to your web-page on gradescope.

Why?

After you have a project idea, it is important to next see what other people have done. This will both save you time (you’ll learn what works and what doesn’t) and prevent you from sinking a ton of effort into something that has already been proved/completed.

For a course project it is fine to repeat what others have done, but your work (code, analysis, etc.) should be your own.

While conducting your literature review you should look out for metrics, datasets, and concepts used in other projects.

It is also an important technique to learn. Reading research papers (and blog posts) is quite a bit different than reading text books. You should not start reading a research paper thinking that you need to understand every concept.

How?

I recommend the following approach for finding research papers:

  1. Write a short list of search terms and synonyms.
  2. Search for these terms using Google Scholar.
    • Look for “recent” papers (for deep learning I’d look for things published in the past two or three years)
    • Focus on papers with higher citation counts, but don’t eliminate a paper just due to the number of citations
    • Look for “survey” and “review” articles, and then click on the “Cited by DDD” link to find recent papers that cite the survey/review article
    • If you find a particularly good article, you can also look at the papers it references (found at the end of the paper)
  3. Search with the same terms using your search engine of choice
    • You can frequently find related blog posts or videos on the authors’ websites
  4. Use a tool like Connected Papers to find related articles.

For other types of work (blog posts, etc.) use your normal searching skills (DDG, YouTube, etc.).

How to Read Articles

Keep the following in mind when reading any research article:

  • Who are the authors? The authors are the only people with the full context needed to understand a paper. Many authors are not native English writers, so don’t let lower-quality writing prevent you from understanding what could be an important takeaway.

  • What is the venue? Every venue will have its own requirements and expectations. Keep this in mind when thinking about what the authors left out (due to space requirements) or didn’t include (due to the media).

  • Who is the audience? You are likely not the intended audience for most research papers. The intended audience is probably some niche sub-field of researchers from around the world. Every group of researchers has its own jargon and informal (not codified) rules and conventions.

In general, I only gripe about grammar and style when reviewing (providing feedback prior to publication) an article, and not when reading an article or reviewing a research proposal. In the latter cases, do your best to look past writing characteristics (good or bad) and focus on contributions.

Once you find a research article (or research heavy blog post), I recommend the following technique:

  1. Skip the abstract
  2. Read the first sentence of every paragraph in the paper
  3. Come up with two or three questions you would ask the author
  4. Summarize paper’s main points in one to three sentences
  5. Read the abstract and compare it to your summary
  6. Read the paper again, but now without skipping any sections.

In general, you only need to read in more detail if you need to understand something in more detail (for example, if you are trying to replicate their work). You should not expect to understand every part of a paper. You do not have the same background or context as the author(s). You will often want to read a paper several times if you are trying to recreate something the authors discovered.

Division of Labor

Every group member should find at least one work to add to your literature review. I do not require every group member to read every cited work. If you have three or four group members, I recommend shooting for six or more cited works, but the requirements are the same for all groups.