Every Binary Code Can Be Realized by Convex Sets

TL;DR

This paper proves that any binary code can be realized by convex sets in some Euclidean space, regardless of whether the sets are open or closed, expanding the understanding of convex neural codes.

Contribution

It demonstrates that all binary codes can be realized by convex sets without restrictions on openness or closedness, using a construction in ^{k-1} for codes with k nonempty codewords.

Findings

Every binary code can be realized by convex sets in ^{k-1}.

The dimension of the realization cannot generally be reduced.

Supports the standard restriction to open or closed convex receptive fields.

Abstract

Much work has been done to identify which binary codes can be represented by collections of open convex or closed convex sets. While not all binary codes can be realized by such sets, here we prove that every binary code can be realized by convex sets when there is no restriction on whether the sets are all open or closed. We achieve this by constructing a convex realization for an arbitrary code with $k$ nonempty codewords in $\mathbb{R}^{k-1}$. This result justifies the usual restriction of the definition of convex neural codes to include only those that can be realized by receptive fields that are all either open convex or closed convex. We also show that the dimension of our construction cannot in general be lowered.