Project

Communication Track

Background & Context

Establish your audience, what you intend to create, why it matters, and what already exists in that space.

Goals

Describe what you intend to create.

Methods

Explain how you’ll make your design choices and build the artifact.

Results

Explain the artifact and its affordances and drawbacks.

Implications

Address what the work implies going forward.

Model-Building Track

Background & Context

State what problem you’re solving, why it matters, and what has already been tried.

Goals

State what you’re trying to achieve — likely a benchmarking target.

Methods

Name the algorithms and architectures you plan to explore.

Results

State what you hypothesize will happen under the methods you’ve described; you may offer competing hypotheses.

Implications

Address what the work implies going forward.

Communication Track

This might be a link to a draft of your artifact, or an outline of what you plan to include for it.

Model-Building Track

This might be plots, loss/accuracy values you’re reaching, or other descriptions of your progress.

Communication Track

Writeup: Write a 1500–2000 word experience report describing your artifact. Examples:

• LLMs Unplugged
• Python Tutor

The sections you decide on are flexible — what best showcases your artifact and the choices you made in designing it? Make sure all of the categories from the SIGCSE webpage are addressed (check the “Experience Reports & Tools (ERT)” column):

• Motivation
• Prior and Related Work
• Approach
• Evidence
• Contribution & Impact
• Presentation

For parts such as “Evidence” that would require feedback from users (which was not required for this project), write about what evidence you or someone else could collect in the future to see what an audience thought of it. Write about what kind of feedback you would expect depending on whether your design choices worked as intended.

Presentation: You will have a 10–15 minute slot to give a presentation to your peers highlighting the portions of your artifact and design process that you would like to share. The presentation does not require any specific format or sections, and is intended to showcase what you feel would be most informative and interesting for the audience. Conferences are often not particularly prescriptive about what a presentation must include—it is a fixed amount of time that the author can use in any way they wish to share their work (e.g. SIGCSE presentation guidelines). What would you like the class and the instructor to take away about your artifact and your design choices? If you were to share it at a conference or pitch, what would you tell your audience?

Model-Building Track

Writeup: Write 1500–2000 words for the Related Work, Methods, and Results sections of a paper describing your model.

Related Work

• Characterize the ARC benchmark itself and its significance, and cite published accuracy numbers so that your result has an external reference point.
• Organize prior approaches by idea (e.g., BERT-related, retrieval-augmented, visual) and synthesize what each family got right or wrong — not a list of “Paper A did X, Paper B did Y.”
• Explain the specific gap or design choice your method builds on.

Methods

• Explain your data and how the model handles it: how multiple-choice is cast as a learning problem, input representation, preprocessing, and exact train/val/test splits.
• Specify the architecture(s), key hyperparameters, and training procedure. Explain your reasoning for each nontrivial choice.
• State your experimental design: baselines, ablations, what’s held fixed, the evaluation metric, and how you tested each component.

Results

• Report how your model compares to prior literature with similar approaches.
• Report on the types of errors your model makes — which question types fail, and whether this matches findings in prior work.
• Include at least one plot illustrating a core finding.
• Include at least one table illustrating core metrics.

Presentation: You will have 3–5 minutes to give a lightning talk to your peers highlighting portions of your model and findings that you would like to share. The presentation does not require any specific format or sections, and is intended to showcase what you feel would be most informative and interesting for the audience. Example guidelines that you may encounter in a conference can be found here in the section for NeurIPS Spotlight Presenters, though you do not need to adhere to these particular requirements.