Bhargava Cube--Inspired Quadratic Regularization for Structured Neural Embeddings

TL;DR

This paper introduces a novel neural embedding method inspired by Bhargava cubes, imposing algebraic quadratic constraints to produce interpretable, mathematically structured 3D representations with high accuracy.

Contribution

It is the first to incorporate number-theoretic algebraic constraints into neural embeddings, enhancing interpretability and mathematical consistency.

Findings

Achieved 99.46% accuracy on MNIST

Produced interpretable 3D embeddings clustering by digit

Enforced quadratic constraints via differentiable loss

Abstract

We present a novel approach to neural representation learning that incorporates algebraic constraints inspired by Bhargava cubes from number theory. Traditional deep learning methods learn representations in unstructured latent spaces lacking interpretability and mathematical consistency. Our framework maps input data to constrained 3-dimensional latent spaces where embeddings are regularized to satisfy learned quadratic relationships derived from Bhargava's combinatorial structures. The architecture employs a differentiable auxiliary loss function operating independently of classification objectives, guiding models toward mathematically structured representations. We evaluate on MNIST, achieving 99.46% accuracy while producing interpretable 3D embeddings that naturally cluster by digit class and satisfy learned quadratic constraints. Unlike existing manifold learning approaches requiring explicit geometric supervision, our method imposes weak algebraic priors through differentiable constraints, ensuring compatibility with standard optimization. This represents the first application of number-theoretic constructs to neural representation learning, establishing a foundation for incorporating structured mathematical priors in neural networks.