The approaches clause explains that the choice of learning method depends on the data available, especially whether it carries labels, and on the task.
What the base framework establishes
The clause opens the approaches group by linking the method to the data. Supervised learning needs labelled examples, unsupervised learning works without them, and semi-supervised, self-supervised, reinforcement and transfer learning each change the relationship between data, feedback and the task. The framework treats the choice as a design decision with consequences for what can be evaluated and what evidence can be produced. The public sample does not carry this clause, so this page describes the group's logic from the public contents and the framework's structure rather than putting words into the clause.
Why it matters to an AI management system
The approach decides what evidence is possible. A supervised model can be tested against held-out labels; an unsupervised model cannot be judged by the same measure; a reinforcement agent is evaluated on reward and behaviour over time. Governance fails when it asks for the wrong evidence because nobody recorded the approach.
The public draft summary names no change to 7.1 General. The final text could still touch it.
Evidence a reviewer samples
| Category | Artefact |
|---|---|
| Approach record | The learning approach, the label situation and the feedback signal. |
| Evidence plan | The evaluation method that fits the approach and what it cannot prove. |
| Alternative note | The other approaches considered and why they were rejected. |
One question to prepare for. What kind of learning is this, and what evidence does that make possible?
Common gaps
- The approach is implied by the library used and never recorded.
- An unsupervised model is evaluated with a supervised accuracy figure that means nothing.
- A reinforcement system is signed off without a record of its reward and its failure modes.
Prepare before publication
Add an approach field to the model card and an evidence plan that matches it. Record the alternatives in one line.
Dates that matter
No separate date attaches to 7.1; the record dates are the stage date, the planning estimate and the CEN expectation stated above.
Cross-references
6.2 Task · 7.2 Supervised machine learning · 7.6 Reinforcement machine learning · 8.4 Modelling
Source
Source for 7.1: ISO/IEC 23053:2022, clause 7.1. The public sample stops before 7.1, so this page relies on the public contents and the public summary and does not quote the clause. The two official records are linked on this page so any line can be checked.
Not legal advice; does not certify. A reading of the framework and the evidence a reviewer commonly samples, not a substitute for the standard or your auditor's rules.