CYBE.PRESS Сигнал безопасности

FIELD NOTE · CYBE.PRESS

What universal approximation does — and does not — promise

The theorem describes what certain neural networks can represent under stated conditions. It does not guarantee efficient learning, good data or reliable behavior.

Technology / systems07 min
01

Representation is an existence claim

Under stated conditions, a network can approximate a broad class of functions. The result says that suitable parameters exist, not that training will find them efficiently.

02

Finite data changes the problem

Real models learn from samples. Generalization depends on the data distribution, inductive bias, optimization and evaluation — none is supplied by approximation capacity alone.

03

Use the theorem at the right altitude

It is a foundation for thinking about expressiveness, not a certificate for a particular architecture, dataset or deployment.