Overview
This advanced course examines the mathematical, computational, and practical foundations of modern machine-learning model development. Topics include supervised and unsupervised learning, regularization, ensemble methods, gradient boosting, kernel methods, neural networks, representation learning, dimensionality reduction, clustering, anomaly detection, imbalanced classification, and selected sequence, text, image, or time-series models.
Students develop end-to-end workflows for problem formulation, feature engineering, data splitting, cross-validation, hyperparameter optimization, pipeline construction, leakage prevention, calibration, threshold selection, cost-sensitive evaluation, and robustness testing. Emphasis is placed on optimization, loss functions, gradient-based training, generalization, bias-variance tradeoffs, and computational complexity.
The course also addresses model interpretation, uncertainty estimation, fairness, subgroup performance, privacy, explainability, distribution shift, model cards, governance, and responsible communication of limitations. Practical work uses contemporary programming and machine-learning libraries, with emphasis on versioned code, reproducible experiments, rigorous comparison of models, and technical justification of decisions in relation to domain objectives.
Learning Outcomes
- Evaluate the mathematical and computational foundations of advanced machine-learning algorithms, including optimization procedures, loss functions, generalization, and computational complexity.
- Select and implement appropriate supervised and unsupervised learning methods for defined analytical and domain objectives.
- Design reliable data-processing, validation, hyperparameter-search, and model-comparison strategies that prevent leakage and support reproducible inference.
- Diagnose overfitting, underfitting, instability, distribution shift, class imbalance, and other model failure modes using appropriate empirical and theoretical evidence.
- Optimize and compare models using feature engineering, regularization, ensemble methods, representation learning, and task-specific evaluation criteria.
- Interpret model predictions using feature attribution, partial dependence, counterfactual reasoning, and uncertainty estimation while identifying limitations of each approach.
- Assess model performance across relevant subgroups and operating conditions using fairness, calibration, robustness, and cost-sensitive metrics.
- Synthesize a reproducible model-development study incorporating versioned code, documented experiments, responsible-AI analysis, and a technically defensible report.
- Defend model-selection and deployment recommendations in relation to domain requirements, risks, governance obligations, and documented evidence.
Timetable
| Type | Length | Frequency | Period |
|---|---|---|---|
| Lecture | 2 hours | Weekly | All semester |
| Lab | 2 hours | Weekly | All semester |
| Tutorial | 1 hour | Fortnightly | All semester |
| Workshop | 2 hours | Fortnightly | All semester |
Assessment Schedule
| Type | Description | Weighting |
|---|---|---|
| Assignment | Model formulation and reproducible data pipeline report | 15.00% |
| Assignment | Advanced methods implementation and analysis assignment | 15.00% |
| Deliverable | Experimental design and evaluation report | 15.00% |
| Test | Practical machine-learning test | 15.00% |
| Capstone | Model development capstone project and technical defense | 25.00% |
| Exam | Final examination | 15.00% |
Prerequisites
- Requirement A university-level course in linear algebra and multivariable calculus
Teaching Staff & Programs
This course is delivered jointly by faculty from the participating programs listed below. In line with the Douchewater Way, the University of Sexology tailors core instruction directly to each cohort's specific discipline — adapting curriculum to program needs rather than forcing students into a one-size-fits-all model. Learn more about our approach at The Douchewater Way.