A neural network is an architecture of connected layers that transforms an input step by step, adjusting the connections during training until the final layer produces the desired output. Canada’s Directive on Automated Decision-Making names it directly, listing “neural networks” among the techniques an automated decision system may use, alongside regression and machine learning.
ISED’s AI advantage blueprint credits the modern version of this architecture to research with deep Canadian roots: the 2018 ACM A.M. Turing Award recognized work that made “deep neural networks a key component of modern engineering,” and, in ISED’s own words, “This breakthrough has led to major advances in computer vision, speech recognition, natural language processing and robotics and many other applications”. Four very different application areas, one shared underlying architecture.
Canada’s Directive on Automated Decision-Making’s Appendix A lists neural networks as a technique distinct from plain machine learning and from deep learning in the same sentence — the government’s own definition treats a neural network as the architecture, and deep learning as what you get when that architecture is stacked into many layers. Statistics Canada’s Q2 2026 survey of AI use by Canadian businesses’s Q2 2026 survey names “Neural networks” as its own, separate application category too, reported by 2.3% of AI-using Canadian businesses (2.5% a year earlier) — second-lowest of the sixteen named categories, ahead of only biometrics (1.8%), and far below “machine learning” (18.2%) or “deep learning” (13.1%). Most businesses using the same underlying architecture evidently describe it under one of those broader labels instead.
An intake scanner built on a neural network does not decide what a document is in one step. An early layer responds to edges and simple shapes in the image; a middle layer combines those into shapes resembling letters and logos; a final layer combines those into a classification — invoice, application form, ID document. Nobody hand-writes the rules connecting one layer to the next; the connections are set during training and then held fixed when the system runs on a new document.
See also deep learning, machine learning and computer vision.
This is one term in a plain-English glossary on how AI actually works and where it fits in a Canadian business.
