# Meaningful neighborhoods and the unclassified remainder

ID: semantic-neighborhoods
Canonical: https://rezonansapp.com/nodes/semantic-neighborhoods
Version: 1
SHA256: 31e9087dea27c11e77c472d30d5efaa1f535144491da7bcb2fd698b3d718c1bd
Epistemic kind: reconstruction

## Question

What counts as a meaningful region of a representation space, and what remains outside the regions a system can recognize?

## Formulation

Face, person, and expression are proposed as overlapping meaning regions. Narrower categories can increase discrimination, while the status of noise or meaningless dimensions remains unsettled.

## Tension

An interpretable region need not be a mathematical manifold; a classifier covering the space need not make everything meaningful.

## Challenge

How could a system encounter a case for which its existing meaning regions are inadequate?