Representation Is Relative: Objects as Invariants Across Readout Mechanisms, Not Points in Embedding Space

Read full text Summarise with VIRAT
Authors: Gabriel Aubin-Moreau
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Using a lattice memory substrate with four readout mechanisms and zero training on CIFAR-10, we demonstrate that visual representation is inherently relative to the readout mechanism. No single embedding captures all category structure. Ship wants value readout (54%), cat wants coupling readout (12%), airplane wants magnitude readout (42%). Each mechanism defines a valid local coordinate system, but no global embedding captures all categories. We propose that an object is not a point in a space but an invariant across readout mechanisms — the pattern of responses that remains consistent when the probe changes. This reframes visual recognition from vector matching to constraint detection across multiple projections.

Fields of Study & Subjects

Advanced Neural Network Applications Domain Adaptation and Few-Shot Learning Explainable Artificial Intelligence (XAI)

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN
Document type
Preprint