CS 601 Project 0
A key aspect of academic work is being able to read and write
scientific papers.
Project 0 is to build an annotated bibliography of selected research
papers.
February 2018
Su Mo Tu We Th Fr Sa
11 12 13 14 15 16 17
18 19 20 21 22 23 24 <- Topics due (in class)
25 26 27 28
March 2018
Su Mo Tu We Th Fr Sa
1 2 3
4 5 6 7 8 9 10 <- Rough paper selections due
11 12 13 14 15 16 17 <- Spring break
18 19 20 21 22 23 24 <- Final writeup due
The papers should relate to a single important research topic.
- Double-counting is highly encouraged. Ideally you'll be able to pick a research topic directly related to your master's project, to other class projects, to a related subject you're interested in, or all three.
- The ideal topic scope covers everything you'll need about the subject: not so broad (e.g., "Machine Learning") that you can't get a good sampling of current work; but not so specific that there are only a few good papers on it (e.g., "Using a 32-16-4-1 backpropagation Neural Network to evaluate checkers boards").
Pick out at least ten relevant research papers.
- I usually start by using an academic search engine, such as Google Scholar, Microsoft Academic, or a general web search.
- Papers in the ACM or IEEE digital libraries can be accessed from on campus. For papers locked away behind a paywall, if I can't find a PDF elsewhere I rarely will pay just to see the paper. (It's frustrating to spend good money for access to what turns out to be irrelevant or poorly constructed papers.)
- Be sure to collect full citations as you gather the papers. (It's often difficult to reconstruct the journal, year, and URL from just the PDF.)
- Citations are like hyperlinks, but designed to keep working after several hundred years.
- I'm OK with technical or survey work in journal papers, conference papers, or book chapters; but not blog posts, Wikipedia articles, or popular magazine articles (these can have useful citations though!).
- A very important source of links to good papers is citations from other good papers. Reading the cited papers is often the only way to understand a paper anyway.
- Old papers are not always useless (Turing's paper on Turing Machines in the 1930's is still completely relevant today), but the change in constants over time often makes possible completely different approaches to the subject (much computer graphics in the 1970's was contorted by the lack of RAM to store a complete raster image, and hence required images to be constructed line-by-line as they were recorded to film).
- The topic scope is expected to drift somewhat during the search for good papers, but try to keep the topic reasonably focused, while maintaining enough diversity between authors and approaches to cover the topic fairly.
Write up an annotated bibliography summarizing the papers.- This would be the "Introduction", "Prior Work", and "References" sections of an actual paper.
- My preferred format in "Prior Work" is about one paragraph per paper, summarizing what the paper contributed to the field:
- A more modern approach to nonlinear iterated function system rendering is to use GPU metaprogramming to estimate the intensity arriving at a given output pixel [Lawlor 2011]. This "backwards" rendering approach has less sampling error than forward projection methods, resulting in smoother images, and also allows multiresolution imaging, allowing much higher resolutions. The primary drawback is the need for GPU-computable analytic inverse functions, and determinants, for each of the geometric distortion functions used.
- In an actual paper, you'd be describing your work on some subject, so you often need to trim down prior work significantly, to a sentence per paper or less. (Competitive academic papers almost always are pushing the page limit to fit in everything you accomplished.)