Over the past few months, I’ve been working on a new library and research paper that unify structure-preserving matrix transformations within a high-dimensional framework (hypersphere and hypercubes).
Today I’m excited to share: MatrixTransformer—a Python library and paper built around a 16-dimensional decision hypercube that enables smooth, interpretable transitions between matrix types like
Symmetric
Hermitian
Toeplitz
Positive Definite
Diagonal
Sparse
...and many more
It is a lightweight, structure-preserving transformer designed to operate directly in 2D and nD matrix space, focusing on:
Symbolic & geometric planning
Matrix-space transitions (like high-dimensional grid reasoning)
Reversible transformation logic
Compatible with standard Python + NumPy
It simulates transformations without traditional training—more akin to procedural cognition than deep nets.
What’s Inside: A unified interface for transforming matrices while preserving structure
Interpolation paths between matrix classes (balancing energy & structure)
Benchmark scripts from the paper
Extensible design—add your own matrix rules/types
Use cases in ML regularization and quantum-inspired computation
Links: Paper: https://zenodo.org/records/15867279 Code: https://github.com/fikayoAy/MatrixTransformer Related: [quantum_accel]—a quantum-inspired framework evolved with the MatrixTransformer framework link: fikayoAy/quantum_accel
If you’re working in machine learning, numerical methods, symbolic AI, or quantum simulation, I’d love your feedback. Feel free to open issues, contribute, or share ideas.
Thanks for reading!