Project
For a detailed explanation of the project, please refer to the Research Project Guide .
Project-group signup is available in the project signup sheet .
Getting Started
Project groups may be the same as your paper-discussion groups or different from them. Aim for a 4-person team whenever possible; one or two 5-person teams are also fine if needed. Each team should choose a memorable team name, and students are encouraged to make that name distinctive and a little ambitious.
Scope
The project should investigate a concrete systems or hardware solution to optimize an AI workload. A strong project identifies a bottleneck, proposes a change, and evaluates whether that change improves a clearly defined metric such as throughput, latency, utilization, energy, memory footprint, or cost.
Proposal
The proposal should define the problem, motivation, research question or hypothesis, primary metric, proposed approach, evaluation methodology, feasibility, risks, related work, and team responsibilities. Teams should be concrete enough to show that the project is doable within the semester.
Midterm and Final
By the mid-semester review, teams should already have a working baseline and initial characterization results. The final presentation and report should tell a clear research story: problem, observation, design, evaluation, limitations, and conclusion.
Timeline
Tue, Aug 25, 2026
Project expectations, example topics, available infrastructure, and team formation will be discussed in class.
Thu, Aug 27, 2026 - Thu, Sep 10, 2026
Read related papers, identify candidate problems, explore the available hardware/software stack, form your team, and settle on a team name.
Thu, Sep 17, 2026
Submit a one-page, single-column proposal. A concise bullet-point summary of the problem, the current state of the art, the proposed solution, and the primary metric is sufficient, as long as it clearly communicates the planned direction.
Fri, Sep 18, 2026 - Thu, Sep 24, 2026
The instructor will provide feedback. Teams may be asked to narrow, expand, or revise the project. All projects must be approved before moving into full implementation.
Tue, Oct 20, 2026 or Thu, Oct 22, 2026
Present concrete preliminary results. Teams should show that the baseline works, explain what has been measured so far, identify the current bottleneck, and describe the remaining implementation work.
Tue, Dec 8, 2026
Present the final project as a poster presentation in Siebel for the broader community to see. There will be no oral presentations for this event. Submit the complete final report together with the project artifacts.
Deliverables
Teams should expect to produce a proposal, a mid-semester review, a final poster presentation, a concise final report, and project artifacts that make the evaluation easy to understand.
Evaluation bar
Strong projects define a meaningful question, choose the right baselines, and support claims with quantitative evidence. A smaller but rigorous result is better than a broad claim with weak validation.