Build on one of the eight recommended papers below and develop your own algorithm, design, or observation. Your project must go beyond summarizing or reproducing the original paper: explain what is new and provide experimental results or data that support your contribution.

At the end of the semester, prepare a 5-minute presentation on your improvement or extension to the original paper and an 8-page, single-column report.

Project requirements · Final deliverables · Recommended papers

Project Requirements

  1. Identify a research question. Explain the original paper’s approach and identify a limitation, an untested assumption, or a new setting worth investigating.
  2. Make a contribution. Propose a new or improved algorithm, extend a system design, or establish a new observation through measurements and analysis. Clearly distinguish your work from the original paper and its released code.
  3. Validate your contribution. Use experiments or data to show why your design is reasonable or your observation is supported. Compare with the original method or an appropriate baseline under comparable conditions. A reproduction can establish the baseline, but it must lead to a new contribution.

Final Deliverables

5-Minute Presentation

Focus on your improvement or extension to the original paper and the evidence supporting it. Briefly introduce the original work, then explain your research question, contribution, key experimental results, and main conclusion.

A suggested allocation is 1 minute for the background and problem, 2 minutes for your algorithm/design/observation, 1.5 minutes for experimental evidence, and 0.5 minutes for conclusions and limitations.

8-Page Report (Single Column)

Prepare an 8-page report in single-column format. Clearly separate the original paper’s contributions from your own work and cite the papers, code, and datasets used.

Suggested report structure:

  • Introduction and motivation: the research question and why it matters.
  • Background and original paper: the relevant approach, assumptions, and limitation being addressed.
  • Your contribution: the new algorithm, design, or observation, with sufficient detail to understand the work.
  • Experimental methodology: implementation, environment, datasets/workloads, baselines, metrics, and settings needed to reproduce the evaluation.
  • Results and analysis: quantitative comparisons, explanations, and additional experiments that test the central claim.
  • Discussion and conclusion: limitations, applicability, and the main takeaway.

Choose one paper as your starting point and develop a well-motivated extension. Check the repository’s setup and data requirements before choosing the experimental scope.

1. ServeGen — LLM Serving

ServeGen: Workload Characterization and Generation of Large Language Model Serving in Production · NSDI 2026
Paper · Code and data

2. SimAI — Networking for AI

SimAI: Unifying Architecture Design and Performance Tuning for Large-Scale Large Language Model Training with Scalability and Precision · NSDI 2025
Paper · Code

3. Agua — AI for Networking

Agua: A Concept-Based Explainer for Learning-Enabled Systems · SIGCOMM 2025
Paper · Code and artifacts

4. StarCDN — Satellite Communication

StarCDN: Moving Content Delivery Networks to Space · SIGCOMM 2025
Paper · Simulator · SpaceGEN trace generator

5. CellReplay — 5G/6G and Mobile Networks

CellReplay: Towards accurate record-and-replay for cellular networks · NSDI 2025
Paper · Code and traces

6. 5G Mid-Band Landscape — 5G/6G and Mobile Networks

Unveiling the 5G Mid-Band Landscape: From Network Deployment to Performance and Application QoE · SIGCOMM 2024
Paper · Data and analysis artifacts

7. CoActo — Edge Computing

CoActo: CoActive Neural Network Inference Offloading with Fine-grained and Concurrent Execution · MobiSys 2024
Paper · Code

8. Pantheon — Edge Computing

Pantheon: Preemptible Multi-DNN Inference on Mobile Edge GPUs · MobiSys 2024
Paper · Code