grad school applicants
Please read this if you want me to support your application to a graduate program at the UofA.
- Deadlines, rules, and immigration procedures change often, and I don’t keep track of them. See the department’s page on how to apply.
- Funding: the UofA funds all CS graduate students. See financial support and awards.
- Background: you don’t need a CS degree, unless your field is very far from CS. I have worked well with students from the humanities, for example.
How I read your email
I answer every email that reaches my inbox. If I didn’t answer yours, it went to spam. LLM-generated emails often do.
I have limited time and narrow interests, so I look for signs that you are able and driven to do research in areas I work on. I will probably not support you if your email:
- asks for information that is easy to find online;
- is about topics I don’t work on, or says you are open to “anything”;
- doesn’t say what drives you and why I would be a good fit for you;
- looks like a mass email: generic praise of my papers, generic claims about what you will bring to my lab, or requests for a meeting.
What I need from you
I decide only after reading your full application, including the reference letters. But before you apply, email me:
- your CV and transcripts;
- a paper review;
- a data challenge report;
- MSc applicants: a description of your best project. PhD applicants: your best paper.
Best project
One page at most, covering a brief description of the project and:
- Methods: how you solved the problem and evaluated your solution.
- Data: where it came from, how it was collected, and how large it was.
- Results: what they show.
- Lessons: what you learned.
Paper review
Pick a paper from my papers, CMPUT 656, or CMPUT 461/561. Write one page, with one paragraph for each of:
- Summary: the problem, why it matters, and what the results mean.
- Methods: how the authors solved the problem and evaluated their solution.
- Strengths: what you think the authors did well. Are the methods appropriate? Is there enough data? Are the conclusions sound?
- Weaknesses: what you think they could have done better. Don’t list future work.
Data challenge
Do one of these, ideally with an AI assistant. Describe what you did using the same structure as the best project. If you used an AI assistant, include your session log.
- Reproduce the paper “On the Long-Tail Entities in News” (Esquivel et al., ECIR 2017). A sample of the data is fine if you lack the computing resources for all of it.
- Analyze the LMSYS-Chat-1M dataset: classify the conversations by topic, and look for signs of anthropomorphization, sycophancy, and fear-inducing language. Suggest other dimensions worth analyzing, and say why.