Optimization for Machine Learning (Fall 2026)

Overview

Each student completes an individual study of an optimization problem for machine learning, supported by theory, experiments, or both. The final outcome is a conference-format research manuscript. Empirical projects also require a reproducible software artifact; theoretical projects do not require code or a software artifact. The project is worth 45% of the course grade: 10% for the proposal, 5% for the technical checkpoint, 10% for the presentation, and 20% for the final report.

Note: COSC 594 projects may develop a meaningful, rigorous extension of existing work. COSC 650 projects are expected to make a stronger original theoretical or empirical contribution. A paper summary or literature review is not a project.

Project Guideline (PDF)

Project Schedule

The weights below are contributions to the total course grade.

Milestone	Weight	Submission	Due date
Proposal	10%	Problem, research question, and plan	Oct 30
Technical checkpoint	5%	Updated proposal with initial results; code for empirical projects	Nov 13
Presentation	10%	Talk slides	Nov 23 or 30
Final report	20%	Manuscript; reproducible software artifact for empirical projects	Dec 9

Deadlines and submission: Submit each milestone on Canvas by 11:59 pm Eastern Time on its listed due date. Presentations take place on Nov 23 or 30; check Canvas for your assigned date, and submit your slides on that date. The final report is due Dec 9. Late report submissions are not accepted.

Reading Guideline

Project Readings (PDF) - Suggested readings and questions for students who have not yet identified a project.

You may also develop other course-related topics. Choose a problem connected to the course and your research interests, and confirm the scope with the instructor before submitting the proposal.

Read the relevant papers with the following questions in mind:

What optimization problem is being studied, and how does it connect to the lectures?
What are the main results, assumptions, and limitations?
What specific question or extension will your project investigate, and how will you evaluate it?

Proposal and Technical Checkpoint

Proposal: Two pages, excluding references, are recommended. Describe the problem and formulation, research question, related work, baseline, and validation plan.
Technical checkpoint: Add a one-page progress summary to your proposal. Include initial results, correctness checks, and the remaining work plan. Empirical projects also include runnable baseline code; theoretical projects may present derivations, theorem statements, or proof progress.

Presentation Guideline

Plan a 10-minute presentation followed by 2 minutes of Q&A. Approximately 10 slides, excluding transition slides, are recommended. A suggested structure is:

Motivation and question: Introduce the machine learning task, its connection to optimization, and the question your project addresses.
Problem setup and prior work: Explain the formulation, notation, assumptions, and relevant existing methods.
Method and evidence: Present the main idea and one or two central technical results, supported by analysis or experiments. Distinguish prior work from your contribution.
Conclusion and limitations: State what the evidence supports, what remains uncertain, and what you will revise in the final report.

Use readable slides, consistent notation, and clear figures. Explain intuitions before detailed mathematics and prepare to discuss assumptions, design choices, and limitations. Audience questions and feedback contribute to class participation.

Report Writing Guideline

Write your report in LaTeX using the NeurIPS 2025 template with the preprint option, and include your name and affiliation. Six pages, excluding references and appendices, are recommended. Include:

Abstract: Summarize the problem, approach, main result, and significance.
Introduction and related work: Motivate the question, identify the research gap, and state your contribution.
Problem setup: Define the mathematical formulation, assumptions, and evaluation criteria.
Method and main results: Present the algorithm, theorem, or empirical finding that addresses the question.
Derivations and analysis: Explain the reasoning needed to assess correctness; place longer supporting arguments in an appendix if needed.
Experiments and validation: For empirical claims, describe the data, baselines, settings, metrics, variability, and relevant diagnostic checks.
Conclusion: Answer the research question at the strength supported by the evidence and discuss limitations.
Contribution and AI use statements, references, and appendices: Identify your contribution, disclose assistance, and attribute reused ideas, code, data, proofs, and figures.

Revise the report using feedback from the checkpoint and presentation. For empirical projects, submit a repository or code archive that reproduces the central results. Theoretical projects do not require code or a software artifact.

Use of Generative AI

Include an AI Use Statement at each stage, documenting assistance and independent verification, or state that no AI tools were used. For AI-assisted work, include representative prompts and materially used outputs in an appendix, and retain complete interaction records until grading. Follow the full disclosure and submission requirements in the project guideline.