Neural representation

A pattern of neural activity that carries task-relevant information under a specified decoding or explanatory model. Anatomical co-location, similar mean activation and a shared representational code are different claims.

A representation whose relevant information is independent of a particular sensory or output modality. Shared mean activation across modalities is weaker evidence than generalisation of information about the represented content, with confounded stimuli and responses controlled.

Conceptual information accessible through different forms such as words, signs and pictures. Shared conceptual content can coexist with modality-specific input and output pathways; an anatomical overlap alone does not establish identical neural codes.

Consider two response vectors $x=(3,1,3,1)$ and $y=(3,3,1,1)$ over the same four measured locations, relative to a zero baseline. Both are positive everywhere and have mean two. Their centered vectors $x_c=(1,-1,1,-1)$ and $y_c=(1,1,-1,-1)$ satisfy $x_c\cdot y_c=0$. Hence identical positive regional means can accompany orthogonal patterns. This counterexample shows why overlapping regional activation does not imply a shared neural representation. It does not assert that real neural codes are necessarily orthogonal or that unsuccessful decoding proves their independence.

The relationship between a neural activity pattern and information represented by that pattern. Testing a shared neural code requires more than activation in the same anatomical territory: the informative pattern or mapping must generalize under an appropriate comparison.