CS 581, Project Proposals

There are two types of projects in this class: survey papers and research papers. In general, research papers require much more interaction with me (in advance of starting) than survey papers, because setting up a research project that can be successful in a short time period is difficult.

In both cases, please expect that your plan will change through our discussions. You are not expected to know all the answers to these questions now -- but fleshing out your ideas in writing and then discussing with me will help you come up with a feasible and still interesting project!

Instructions for survey papers

• Specify the topic you will survey in enough detail to be clear what you are planning to work on.
• List 10-20 papers that you think might be good for your survey. Make sure you have at least read the abstract for all of them, and preferably glanced through each paper (to see how long it is, whether you think you can read it at least at a high level, etc.). The goal is to have a long enough list that it will probably contain a good set of papers, but most likely we'll whittle it down to a smaller set. It is probably best to mainly focus on recent literature (last 5 or perhaps 10 years), but do include major early contributions if necessary.
• For each paper in your list, write a one paragraph summary of its contributions, why you listed that paper, and how relevant you think it is to your survey.

Research project

• Specify the objective (design new method? compare existing methods? analyze existing dataset using a new method and compare to old analysis?)
• Specify what is known about the problem, and in particular what are the current best methods (back this up with a reference to the literature that supports your statement).
• Specify why there is a value in doing your project.
• If you are comparing existing methods, say which methods and why selected.
• If you are designing a new method, give as much detail about what you are planning to do. You may not know at this time exactly what you will do, but specify as much as you can of what algorithmic designs you will explore.
• If you are analyzing an old dataset (publicly available) with a new method, explain why this new analysis is worth doing.
• For all types of projects, describe the experimental evaluation in as much detail as you can: (a) what datasets you will explore (and what their properties are -- size, heterogeneity, etc.), (b) what criteria you will use, (c) how much time you think this will take, and (d) where you will run the analyses. Make sure you have thought about the computational issues (i.e., do you have the environment you need for this?).
• Provide a bibliography for your project (but here the list may only be 5-10 papers). For each paper in your list, write a one paragraph summary of its contributions, why you listed that paper, and how relevant you think it is to your survey.
• Also, and this should go without saying, for every existing method you will use, provide version numbers and commands.