Network Fundamentals, Learning, and Agentic AI. Explore how networks behave and how models and agents can improve them through hands-on programming and experiments.
Spring 2027
- Instructor
- Xiaochan Xue
- Institution
- University of Hawaiʻi at Mānoa
- Meeting Time
- M/W 9:00–10:15 AM (room TBA)
Credits: 3
Prerequisite: ECE 468 or consent
Format: Project studios and four major programming projects
Syllabus
- Topics, assessment weights, and deadlines may be revised before the semester begins.
Course Description
Study network architecture, switching, protocols, routing, congestion control, and wireless services, then apply prediction, resource allocation, and automation methods to wired, wireless, and edge networks. AI-RAN and ISAC are optional application areas. Official catalog description →
The projects build from measurement to prediction, decision-making, and automation. Students first explain network behavior, then learn to forecast demand and allocate resources. The final project brings these methods to a chosen networking problem, connecting model or agent behavior to measurable system outcomes.
How the Class Works
Eight short lectures of 35–45 minutes, two per project, introduce the core concepts. Class time also includes programming, student-led discussions, project feedback, and demonstrations. Read the project notes and attempt the prelab before starting each project; bring code, results, or questions to studio meetings.
Each studio has a focused task, such as comparing approaches, diagnosing an unexpected result, or choosing the next experiment. Students work with the project code and discuss evidence from plots, traces, and implementation choices, with feedback from peers and the instructor.
Learning Outcomes
By the end of the course, students should be able to:
- Explain how protocols, routing, queues, and resource constraints affect network performance.
- Design reproducible experiments using throughput, delay, loss, and fairness measurements.
- Train predictive models and evaluate them on unseen network conditions.
- Compare rule-based, optimization-based, and learned control policies.
- Build a network application or agent and explain its benefits and limitations.
Proposed Assessment
| Project | Target window | Weight |
|---|---|---|
| 1 · Network Behavior and Performance | Weeks 1–3 | 20% |
| 2 · Learning from Network Data | Weeks 4–6 | 20% |
| 3 · Programmable Networks and Intelligent Decisions | Weeks 7–10 | 25% |
| 4 · Open-Ended Network Intelligence Project | Weeks 11–16 | 35% |
Project 4 includes the final demonstration and report; there is no separate written final exam. Prelabs, progress checks, and in-class activities are part of these four assignments. Collaboration rules, late policy, and deadlines will be confirmed before classes begin.
Assessment emphasizes networking understanding, correct implementation, controlled comparisons, and clear interpretation. Explain design choices, document reproducible experiments, and investigate failures. Model complexity or performance gains alone do not determine the grade; well-supported negative results can also demonstrate strong understanding.
Preparation
Familiarity with Python and basic probability is helpful. Prelabs introduce project-specific tools, with more guidance in the early assignments and greater choice in the final project.
Projects
- The outlines below describe the planned projects. Detailed requirements will accompany each project release.
Project 1 · Network Behavior and Performance
Is the shortest path always the fastest? Investigate how traffic and link conditions move the bottleneck.
- Starter: a two-path queue simulator with routing baselines, packet traces, and plots.
- Task: implement an adaptive path-selection policy and compare it under normal load, bursts, and degraded links.
- Evidence: throughput, tail delay, and packet loss; explain when the policy helps or fails.
Project 2 · Learning from Network Data
Can a small model predict network demand when traffic patterns change?
- Starter: a synthetic traffic generator, independent data splits, simple predictors, and a trainable linear model.
- Task: choose features and history windows, train a predictor, and compare it with the baselines.
- Evidence: prediction errors on unseen traffic and failure cases, with checks for data leakage.
Project 3 · Programmable Networks and Intelligent Decisions
How should limited resources be shared between delay-sensitive and high-throughput services?
- Starter: a two-class allocation simulator, rule-based policies, Q-learning, and evaluation on held-out episodes.
- Task: build an optimization-based or learned policy and test its behavior across traffic conditions and repeated runs.
- Evidence: throughput, delay, fairness, and resource-constraint checks.
Project 4 · Open-Ended Network Intelligence Project
Choose a networking problem and test what a model, controller, or agent contributes.
- Starter: the default track provides a diagnostic-agent sandbox, expert demonstrations, decision-tree policy training, and tool-trajectory evaluation.
- Task: define a testable question, build or train a method, and compare it with a baseline on held-out scenarios.
- Evidence: network or task performance, resource costs, and failure cases; agent projects also examine tool-use errors and action outcomes.
Alternative topics include routing, scheduling, edge offloading, AI-RAN, ISAC, and network automation. Explore ideas during Weeks 8–9. Confirm data, interfaces, and compute needs in the proposal; GPU resources and live testbed access depend on availability.
What to Submit
Each submission includes runnable code, experiment configurations, key figures, a concise analysis, and a short demonstration or explanation.
Materials
- Course materials: code, project guides, and the report template will be shared through the course GitHub repository. Links to each project will be added when ready.
- Requirements: Python 3.9+, a CPU, and no additional packages. Core experiments use local simulation and synthetic data.
- Optional wireless case studies: OCUDU · OpenISAC-Edge. Follow releases and discussions for possible project ideas and contributions.
Schedule
- Week numbers exclude spring recess; milestones are target weeks, not confirmed deadlines. See the UHM Spring 2027 Academic Calendar.
| Week | Monday | Wednesday | Project milestones |
|---|---|---|---|
| 1 | Course orientation Project roadmap, assessment, student interests, and environment setup |
Short lecture 1/8 Network architecture, paths, and capacity; trace a packet and run the baseline |
Project 1 introduction and setup |
| 2 | No class · MLK Day | Short lecture 2/8 Queues, congestion, delay, and loss; compare buffer sizes |
Project 1 baseline check |
| 3 | Programming studio Implement and compare adaptive path policies |
Student results discussion Project 1 demonstrations, bottlenecks, and failure cases |
Project 1 results and submission target |
| 4 | Short lecture 3/8 Telemetry, causal features, and data splits; audit a prediction window |
Data studio Inspect episodes, reproduce simple predictors, and check leakage |
Project 2 introduction and data preparation |
| 5 | Short lecture 4/8 Small-model training and validation; fit and reload a predictor |
Model clinic Feature experiments, validation choices, and paired comparisons |
Project 2 training and progress check |
| 6 | No class · Presidents' Day | Student results discussion Project 2 demonstrations, traffic shifts, and forecast failures |
Project 2 results and submission target |
| 7 | Short lecture 5/8 Resource allocation, deadlines, and fairness; step through an episode |
Programming studio Compare equal-share and demand-based allocation |
Project 3 introduction and rule-based baselines |
| 8 | Short lecture 6/8 Q-learning; inspect states, actions, rewards, and policy objectives |
Training studio Train a policy and vary exploration or state design |
Project 3 implementation; explore Project 4 topics |
| 9 | Student-led case discussion Map the allocation model to wireless resources and RAN architecture |
Case exploration and project clinic Students examine AI-RAN, OCUDU, or OpenISAC-Edge and discuss feasible extensions |
Project 3 progress check; discuss Project 4 feasibility |
| — | Spring Recess (Mar 15–19) · No class | ||
| 10 | Evaluation studio Repeat training and compare policies under changed traffic |
Student results discussion Project 3 demonstrations and resource-allocation tradeoffs |
Project 3 results and submission target |
| 11 | Short lecture 7/8 Agent observations, tools, and task success; inspect a diagnostic trajectory |
Proposal workshop Student pitches, peer feedback, and evaluation-plan review |
Project 4 proposal and evaluation plan |
| 12 | Short lecture 8/8 Expert demonstrations, training, and held-out evaluation; train a tool policy |
Training and integration studio Build the chosen model, controller, or agent |
Project 4 baseline and initial implementation |
| 13 | Project clinic Debug workflows and review initial failures in small groups |
Midpoint demonstrations Students show a working system, early results, and one unresolved issue |
Project 4 progress check |
| 14 | Student-led topic exchange Share project-relevant findings on edge computing, model selection, or another chosen topic |
Robustness studio Run changed conditions, ablations, and failure analysis |
Project 4 robustness and comparison experiments |
| 15 | Peer reproduction studio Run another group's small example and prepare the final demonstration |
Final demonstrations Student presentations and peer discussion |
Project 4 demonstrations begin |
| 16 | Final demonstrations Student presentations and evaluation discussion |
Project retrospective Student reflections, reproducibility handoff, and a brief instructor wrap-up · Last day of instruction |
Project 4 final demonstration and report target |
Other Academic Dates
- Mar 26: Prince Kūhiō Day and Good Friday, non-instructional day.
- May 6-7: Study period.
- May 10-14: Final examination period.
- May 14: Semester ends.