Resources
Understanding machine learning
A model learns a rule from examples, then applies it to new cases. The whole craft sits in three questions: which examples, which rule, and how to verify it holds beyond the training data.
Learn a rule, verify it elsewhere
A regression model estimates a quantity, a classification model assigns a category. In both cases it adjusts its parameters to reproduce the supplied examples as closely as possible. The decisive question comes next: does that rule hold on data it has never seen? Cross-validation answers by holding back part of the examples for evaluation. A model that shines on its training set and degrades on the test set has memorised rather than learned, and the protocol makes that diagnosis immediate.
- Examples representative of the situation the model will serve
- A strict separation between training data and test data
- A performance measure chosen for the target decision
- A comparison against a simple rule, used as a baseline
- An interpretation of the variables that weigh, required on any decision
The five decisions of a modelling project
- 01
Define the target
What the model must estimate, in the decision's own unit. A vague target produces an unusable model.
- 02
Assemble the examples
A representative period, a stated scope. This is where everything else's validity is decided.
- 03
Set a simple baseline
The mean, the last known value, a business rule. The model must beat that to justify itself.
- 04
Compare several families
Linear models, trees, kernel methods: cross-validation decides on an identical protocol.
- 05
Explain before deploying
The variables that weigh are presented to the business. A model you can explain is a model you dare to use.
The programmes that carry modelling
Two Bac+5 level programmes, two angles: model design and industrialisation on one side, applied statistics and data engineering on the other. Taught volumes compare directly.
| Programme | Level | Duration | Credits and hours | Fees |
|---|---|---|---|---|
| Artificial Intelligence programme — Bac+5 level | Master's degree | 2 years | 120 credits · 1200 h | 7 800 000 GNF per year |
| Data Science programme — Bac+5 level | Master's degree | 2 years | 120 credits · 1200 h | 7 200 000 GNF per year |
Questions about models
How much data is needed?
Enough examples to cover the situations the model will meet, which depends on the problem's variety far more than on a universal threshold. A well-built simple rule serves usefully in the meantime, and it is the baseline to beat.
Do I need mathematics to model?
The programme devotes 220 taught hours to linear algebra, calculus, probability and optimisation, taught for the use modelling makes of them. It is the block that makes the following ones legible.
How do I explain a decision produced by a model?
By presenting the variables that determined it and their meaning for the business. Interpretability has a whole block in the data science programme: it governs a model's acceptance far more than its raw performance.
Explore next
- Models in productionInference service, drift monitoring, retraining, running cost and the trade-off between local and cloud processing.
- Data governanceData ownership, life cycle, hosting, retention, reversibility and vendor dependence: framing governance.
- Personal dataPurpose, legal basis, minimisation, retention, security and individual rights: the processing register in practice.
- Industrial site safetyHSE management system, work permits, co-activity, internal audit, leading indicators and safety culture: a site's complete arrangement.
- Assessing a riskHazard, exposure, criticality, risk register and the hierarchy of prevention measures: risk assessment applied to a real workstation.
- Incident investigationFact gathering, causal tree, root causes, action plan and lessons learned: conducting an investigation after an incident or a near miss.