School of Computational Arts & Sciences Graduate 6 ECTS
AI520 · Responsible ML Systems
Design and operate responsible ML services: evaluation protocols, bias and slice checks, monitoring/alerts, rollback strategies, and incident-style postmortems. Emphasis on documentation, governance, and shipping models that remain reliable under drift.
- ai
- ethics
- mlops
| Overview | Details | Notes |
|---|---|---|
| Code | AI520 | — |
| Title | Responsible ML Systems | — |
| School | School of Computational Arts & Sciences | — |
| Level | Graduate | — |
| Credits | 6 ECTS | — |
What you will learn
- · Design evaluation protocols for deployed ML systems
- · Implement monitoring and incident response playbooks
- · Document model behavior with audit-friendly artifacts
Prerequisites
- · Programming experience
- · Basic probability and statistics
Assessment
| Component | Weight |
|---|---|
| Evaluation memo | 25% |
| System labs | 35% |
| Incident postmortem simulation | 15% |
| Final project (deployment plan + review) | 25% |
Weekly outline
Week 1: Failure modes that matter
3 topics
- · What counts as harm
- · Stakeholders
- · Boundaries and constraints
Week 2: Baselines and evaluation
3 topics
- · Datasets and drift
- · Metrics vs. values
- · Reproducible protocols
Week 3: Monitoring and logging
3 topics
- · Telemetry
- · Alert fatigue
- · SLOs for ML
Week 4: Governance
3 topics
- · Documentation
- · Review workflows
- · Change control
Week 5: Incident response
3 topics
- · Runbooks
- · Rollback
- · Retirement plans