Skip to content

Assignments

Grading

Component Weight
Assignments (5) 50%
Course Project 30%
In-Class Paper Quizzes 15%
Participation 5%

Letter grades follow the overall percentage: A ≥ 90, B 80–89, C 70–79, D 60–69, F < 60.

Assignment list

There are five individual assignments, each released and due as shown below (all due at 12:00 pm / noon). See the Schedule for the full week-by-week plan.

# Title Released Due Notebook
HW1 Tokenization & N-Gram Language Models Jan 27 Feb 10 Tokenization & Vocabulary
HW2 Machine Learning & Text Classification Feb 17 Feb 26 Classification Fundamentals
HW3 Neural Text Classification & Structured Prediction Mar 3 Mar 24 Embeddings & Sequence Models
HW4 Neural & Pretrained Models (RNN/LSTM, Transformer, BERT, T5) Mar 31 Apr 9 Transformers & LLMs
HW5 LLM Post-Training & Agentic AI Apr 14 Apr 23 SFT with LoRA/QLoRA · Agentic NLP

Each assignment is a runnable notebook. Open it from the Table of Contents to read it inline or launch it in Colab.

Course project

The final project is done in groups of at most 2 (individual projects require prior approval). You may pick from a list of shared tasks or propose your own research project. Data and evaluation methods must be clearly defined, accessible, and ready to use. No double-dipping projects across classes. The project is graded across three milestones:

  1. Proposal (Week 4) — initial project proposal.
  2. Progress report & mid-term demo (Week 10) — 10-minute demo with preliminary results, during office hours.
  3. Final report & presentation (Week 15) — final report plus an in-class presentation with live Q&A.

Submission policy

  • Assignments are submitted through Canvas; programming work is in Python.
  • Include both your report and code.
  • The Canvas timestamp is the official submission time.

Late policy

Assignments are due at 12:00 pm (noon). Late work is accepted up to 24 hours after the deadline with a 10% penalty (e.g., a 90 becomes 81). Nothing is accepted more than 24 hours late. This is enforced strictly — a 12:01 pm submission incurs the 10% penalty. Documented extenuating circumstances (e.g., a medical condition with University-issued documentation) will be handled case by case.

Quizzes

  • Scheduled quizzes (counting toward the 15%): ~20–30 minutes, 5–10 questions, in class on paper, closed book (paper notes allowed). Scope and date announced at least one week in advance.
  • Unannounced check-ins (participation only): 2–3 minutes, 1–2 questions. Bring a device to answer quickly.

Generative AI policy

You may use free generative-AI tools (e.g., Copilot, open LLMs) for assignments and the project, but you must not fully delegate your code to them — prefer autocompletion over wholesale generation. You must cite any AI usage (tool, version/date, and what it produced) and include an AI-usage reflection statement at the end of each assignment, written without AI. No AI is allowed on in-class quizzes. See the Syllabus for full details.

Academic integrity

  • You may discuss ideas, but may not look at or share code — similar code scores zero. Document anyone you discussed solutions with.
  • Cite all external resources (papers, libraries, Discord, StackOverflow, AI tools).
  • Programming work is checked with collaboration-detection software. Violations are reported per the OU Academic Misconduct Code.