Applied Machine Learning
Model Evaluation Clinic
Calibration, slice analysis, and human-in-the-loop rubrics for models that must ship with defensible metrics.
- Duration
- 3 weeks · online
- Format
- Evening sessions
- Level
- Intermediate
- Tuition (informational)
- ₩650,000
Program narrative
We emphasize honest limitation statements in model cards. You will run slice finders, write calibration narratives, and design lightweight human audit loops that do not overburden reviewers.
What is included
- · Slice discovery notebooks with false-positive archetypes
- · Calibration plots with decision threshold memos
- · Human rubric spreadsheets with inter-rater agreement math
- · Leakage sniffers for temporal splits
- · Fairness metric chooser with caveats section
- · Dry-run of a release review meeting
- · Template for “metrics we will not claim” section
Outcomes you can demo
- · Ship a model card draft your legal partner can comment on
- · Identify two slices where performance is unacceptable
- · Propose a monitoring chart that catches silent drift early
Mentor of record
Amira Haddad
Risk modeling background; leads evaluation clinics for regulated adjacent industries.
Participant questions
Is this a statistics refresher?
We assume working knowledge of precision/recall. We spend little time on textbook derivations and more on operational reporting.
Legal sign-off included?
No. We prepare artifacts your counsel can review; we do not provide legal advice.
What is not included?
Automated fairness remediation beyond detection; that is a separate ethics deep dive.
Recent participant notes
“Model Evaluation Clinic’s slice finder notebooks are now our default attachment for internal model releases.”