Link Search Menu Expand Document
CS 492

Assignment 2: Finding Research Problems

For this assignment, you’ll write another short (2-page) paper describing and brainstorming two (2) open problems in a research area of your choosing.

Part 1: Finding Problems

Pick a research area first (it can be the same or different area from your first assignment). Then you’ll want to look at top conferences/journals in that area. For example, for me, I would look through the paper titles at a recent OSDI iteration and find papers related to two topics that interest me. For example, I might collect papers from top conferences for the past 3 years related to distributed storage. If you don’t know what the top places are, ask a professor in that area!

Once you’ve got several (recent, e.g. published in past 5 years) papers for each topic, your goal is then to use them to find one open research problem for each topic. The problems can be as broad or as focused as you’d like. For example, suppose my chosen topics are “system software for machine learning” and “distributed consensus.” After reading my handful of papers related to those topics at a high level (you don’t have to understand everything), I might have something like this:

####Topic: Better System Software for Machine Learning Problems Identified:

  • Most ML models have too much data movement between GPU and CPU
  • We need to build systems that allow model parameters to span machines
  • Reducing model precision to save memory consumption

####Topic: Improving Distributed Consensus Problems Identified:

  • Existing consensus algorithms are too expensive across different data centers
  • Write-heavy workloads perform poorly on existing coordination schemes
  • Consistency models used for global coordination are too restrictive for programmers

How to find problems from papers

Here are some hints for finding the problems:

  • Try to understand the abstract of the papers first. That will give you an idea of the problem being solved in these papers.
  • Look at the discussion section near the end and the conclusion section. Did they solve the problem completely?
  • Look for challenges the authors encountered in the discussion or evaluation section of the paper. A huge hint is if they write something like “this is not within the scope of this paper; we leave X for future work”. That’s an easy target for you.
  • Are the paper’s performance results significant? Or is there a very small improvement over the state of the art. If so, that might be an open problem!
  • What does the paper’s “future work” section say? What did the authors leave undone or what do they say needs to be done?

After you’ve identified your problems, make a section in your paper for each one and give a brief (2-paragraph) introduction to the problem.

Part 2: Brainstorming

For each problem you found above, write a few paragraphs discussing the following:

  • Why it’s important to solve
  • Why it’s challenging to solve
  • What papers you read related to the problem
  • Any ideas you might have on how to begin solving it.

Submission

Please e-mail me your paper as a PDF. This is due on Monday, December 5, 2022 before class