What problem does this paper attempt to address?

The problem that this paper attempts to solve is to develop a fully tensorial hyper - complex neural network method, enabling neural networks to operate based on arbitrary algebras. Specifically, the core contributions of the paper include: 1. **Representing Algebraic Multiplication as a Third - Order Tensor**: The author observes that algebraic multiplication can be represented as a third - order tensor and uses this tensor to perform calculations in each algebraic operation. This enables neural network libraries to support efficient tensor operations. 2. **Generalization to Algebras of Arbitrary Dimensions**: The algorithm proposed in the paper is not limited to four - dimensional algebras (such as quaternions), but is applicable to algebras of arbitrary dimensions. This is very important for processing data that is naturally encoded as n - tuples. For example, RGB color data can be encoded as a single element in a four - dimensional algebra. 3. **Simplifying and Generalizing Existing Methods**: The paper improves the existing hyper - complex neural network implementations and avoids the complex step of constructing additional multiplication matrices. By directly inputting the multiplication table of the algebra, the algorithm can be applied to any algebraic structure, not just a specific four - dimensional algebra. 4. **Providing General - Purpose Algorithms**: The paper provides general - purpose algorithms for hyper - complex dense layers and convolutional layers. These algorithms can handle algebras of arbitrary dimensions and are consistent with the existing four - dimensional algebra implementations. ### Summary of Main Contributions - **Theoretical Basis**: Proved that algebraic multiplication can be represented as a third - order tensor and provided relevant mathematical derivations. - **Algorithm Implementation**: Proposed general - purpose algorithms for hyper - complex dense layers and convolutional layers applicable to algebras of arbitrary dimensions. - **Application Expansion**: Generalized the existing four - dimensional algebra methods, making them applicable to a wider range of algebraic structures, especially in fields such as image processing and time - series analysis. Through these contributions, the paper aims to enhance the flexibility and applicability of hyper - complex neural networks to handle more types of complex data processing tasks.

Fully tensorial approach to hypercomplex neural networks

KHNNs: hypercomplex neural networks computations via Keras using TensorFlow and PyTorch

Demystifying the Hypercomplex: Inductive Biases in Hypercomplex Deep Learning

Tensorized Hypergraph Neural Networks

Fully Hyperbolic Neural Networks

Towards Explaining Hypercomplex Neural Networks

Easily Scalable Photonic Tensor Core Based on Tunable Units with Single Internal Phase Shifters

Stable tensor neural networks for efficient deep learning

Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods

Convolutions Through the Lens of Tensor Networks

Parameterized Hypercomplex Graph Neural Networks for Graph Classification

NeuLFT: A Novel Approach to Nonlinear Canonical Polyadic Decomposition on High-Dimensional Incomplete Tensors

T-HyperGNNs: Hypergraph Neural Networks via Tensor Representations

Universal Approximation Theorem for Vector- and Hypercomplex-Valued Neural Networks

Gluing Neural Networks Symbolically Through Hyperdimensional Computing

The Tensor as an Informational Resource

Hypercomplex Representation of Finite-Dimensional Unital Archimedean f-Algebras

Fourier Circuits in Neural Networks and Transformers: A Case Study of Modular Arithmetic with Multiple Inputs

Stable Tensor Neural Networks for Rapid Deep Learning

Tensor Programs II: Neural Tangent Kernel for Any Architecture

Deep Complex Networks