Natural Language Processing Capstone

UW CSE 481M, Spring 2025
M/W 1.30pm-02:50pm, MGH 295

Goals of the Course

This class will provide students with an intensive 10-week experience in successfully completing a challenging, well-scoped research project on a topic related to natural language processing and its many applications.

Participants will work in small groups (approximately 3 people in each group) to hone their technical skills to quickly absorb and adapt new technical knowledge, gain experience in complex programming, perform thorough experiments and analysis, and learn how to find a path when faced with negative results.

Additional objectives of this class include:

• technical communication skills to produce high quality interim technical reports that inspire insightful discussion across project groups,
• advisory project experience to provide technical advice and constructive feedback to others, and project management skills to prioritize work items to maximize the chance for successful outcome.
• Though not required, some project options in this class have a benefit to the larger community of NLP researchers and practitioners, through the generation of reproducibility reports and open-source artifacts.

Project Options

There are two options for projects:

1. Reproducibility project (“option R”): reproduce experiments in a recently published NLP paper. Detailed instructions for this option are given
2. Design your own (“option D”). Your team designs a project around something you are interested in. The course staff will do our best to advise you directly or to connect you to local researchers with relevant expertise.

The course calendar details the deadlines for the project’s written reports.

In addition to the written reports, your team is expected to give four five-minute updates over the course of the quarter. Refer to the course calendar for the dates of your team’s updates. You are required to use no more than three slides (including the title slide) for each update. Finally, there will be a final presentation either as a video (which you will upload to Youtube) or as a live poster session at the end of the quarter.

Evaluation

Students will be evaluated as follows:

1. Five-minute updates (short talks given in lecture; four at 5% each, shared by the team)
2. Individual reports; (two at at 2.5% each)
3. Team blog reports on edstem; (five at 5% each, shared by the team)
4. Team formal write-ups; (three at 10% each, shared by the team)
5. Participation (feedback on classmates’ posts and presentations): 10%
6. Final poster presentation: 10%

Your team may use up to three penalty-free late days (total, not per report) on any report other than the final report. Late day on a team report is deducted from each member. Work turned in late after the late days are used up will get a zero grade. Feedback to other teams’ presentations will be tracked on presentation days. Giving feedback to other teams’ reports on Ed is also encouraged since it will lead to more informed questions in class.

Hardware

Two servers are available for class use:

• dashi – 4× NVIDIA GeForce RTX 2080 Ti GPUs
• bonito – 2× NVIDIA GeForce RTX 2080 Ti GPUs

To access the servers, you must be on campus or connected via the Husky OnNet VPN (see Ed for setup instructions).

Use the following SSH commands to connect:

  ssh [cse_netid]@bonito.cs.washington.edu 

  ssh [cse_netid]@dashi.cs.washington.edu  

Calendar

Resources

• EdStem. Course communication will be via EdStem