What the document means by a neural network
The introduction describes neural networks as common AI systems used for detection, classification and interpolation across text, speech, images, time series and tables. Part 1 describes one as a set of simple processing units joined by weighted connections whose weights can be adjusted.
The requirement in plain English
The introduction describes neural networks as common AI systems used for detection, classification and interpolation across text, speech, images, time series and tables. Part 1 describes one as a set of simple processing units joined by weighted connections whose weights can be adjusted.
What the public record says
The introduction lists the task types and data types. Part 1's public term 3.4 gives the technical definition and notes that most neural networks are used as realisations of the connectionist model.
Why it matters to you
The inventory question is harder than it looks. A neural network can arrive inside a bought product, a pre-trained model, a recommender, a forecasting module or a document classifier. The definition is broad enough to include all of them, and the public scope is silent on deployment model, so nothing in it excludes a bought system. The task and data list is a useful checklist for the inventory: if a system does detection, classification or interpolation on any of those data types, ask whether a neural network is inside it.
Evidence a reviewer asks for
- The inventory, one row per system, with task type, data type and supplier.
- The source of the neural network: trained in house, bought, or embedded.
- The owner and the version of each model.
- A note on any system whose architecture is unknown.
Checks to run this month
- Does the inventory cover bought and embedded models?
- Is the architecture known for every row?
- Which systems do detection, classification or interpolation?
Dates
FDIS registered 2026-08-21. Part 1 published as a Technical Report in 2021.
Cross-references
What this page does not state
The public text does not exclude any sector or deployment model.