Research Project Fair

Due (in class) Wednesday, September 21st, 2022

Please describe your research project interests to your classmates in no more than 60 seconds. This will give you a chance to figure out who you might want to work with.

Form Groups

Due Wednesday, September 28th, 2022

Please form groups of two, then email myself and Liam with a vague idea of what your project is. I might allow some groups of three, but expectations will be at least 33.3% higher, and we’ll need to discuss it first.

Proposal

Due Wednesday, October 12th, 2022

Please submit a research proposal using the IEEE latex template (look for ieeeconf.zip).

Your proposal should include the following sections:

abstract-abstract

In a single sentence, what are you studying and how? Very clear ideas can usually be expressed briefly to non-experts. As you write your full paper, you’ll expand this to a paragraph (at which point it will simply be an “abstract”).

justification

Take about a paragraph and explain why your peers should find your work interesting. To some extent, this might be about broad social benefit, but more frequently, this is about your project’s relevance to other research. For example, a project offering a new captioned image dataset will enable other scientists to build better language models for working with images.

Although no one will be counting your citations, you eventually will want 20-30 of them, and most of that discovery will happen earlier in your project. At this point, you should have two to three closely related papers you’re drawing inspiration from. The related work section should not only summarize those papers’ contributions, but also differentiate your work from theirs. You should compare – how is what you’re doing similar to what’s been done? And you should contrast – how does your work differ?

method

In another paragraph, please describe how you plan to solve your problem. A paper about a new language model, for example, would include a mathematical description of that model. A paper about a new dataset would describe how the data was collected. Diagrams are very helpful, generally so but particularly in this section.

data

If you’re using existing data (and you probably should be), please identify the source. Also include a description of the dataset statistics (how many samples? languages?), as well as some examples from the dataset, so that your reader can see some of what you can see. If you’re using a simulator (or live robot), please also describe the simulator in some detail.

evaluation

A very important section, the evaluation tells the reader how you know whether or not you were successful. Most projects will probably have a table that looks similar to this one: results Your metrics might be accuracy, BLEU, MRR, or distance. Please include as much information you can about how performance will be measured.

Midterm Presentations

Due (in class) Monday, November 7th, 2022

  • 2 slides, about 5 minutes.
  • Please tell your classmates your task, evaluation, method, and results so far.
  • Diagrams are great.
  • Examples from your dataset or simulator are very helpful.
  • Please give baseline results (random, naive, and/or a competitor model)

Final Presentations

Due (in class) Monday and Wednesday, December 5th and 7th, 2022

Final Submission

Due Friday, December 16th, 2022

Paper

Formatted as a submission to ICRA. You don’t need to consult the full instructions for authors, but it’s worth taking a look to see what a real submission process looks like.

In general, you are simply filling in what you began in the research proposal. Hopefully you will be able to reuse some of that work. For a full list of sections we expect in your paper, please see that assignment, above. Remember also to use the same latex template linked for the proposal, also above.

There is a 6+n page limit. You may (and should) have up to six pages with n additional pages for your bibliography.

Your paper should have at least three reasonably well made figures, in addition to your results table. These figures are the first thing your reader will see and can illustrate complex topics like your task, your model, or your evaluation method. They don’t need to be beautiful, but if they look bad, it will hurt the perception of your paper.

Include several specific examples of your dataset.

Also include specific results of your model that you found interesting. Usually these are times when your model was particularly good or bad. You might include some of these results as one of your three figures.

Code

A github repository with your code.

Your code should be well documented.

The readme should tell a user how to train and evaluate your model. If a separate dataset is needed, refer us to the instructions for downloading it. Once downloaded, let us know how to link to it from within your repo.

You should include some way of managing user dependencies, like a requirements.yaml file.

You should confirm that a fresh unix machine can clone your github repo, link to your dataset, and train and evaluate your model. Your instructors will be doing this as we grade your work.