Hypercomplex Neural Network Processing Circuit
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning technologies using real numbers in neural networks are limited in their ability to efficiently process complex and quaternion-based data, which is essential for advanced signal processing in medical imaging and other fields, as they lack the operational closure and mathematical richness of hypercomplex numbers.
Innovation Solution
A hypercomplex-number operation device that includes processing circuitry capable of acquiring and processing hypercomplex numbers, applying them through parametric functions with different forms for real and imaginary components, and generating output data, integrated with medical imaging apparatuses like ultrasound and magnetic resonance imaging systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If real numbers are used in neural networks, then the system has completeness and algebraic operations are convenient, but the ability to efficiently process complex and quaternion-based data is limited
Solution Approach 1:
The patent changes the fundamental parameter of the number system from real numbers to hypercomplex numbers (complex numbers or quaternions). This allows the neural network to natively process complex and quaternion-based data without conversion, directly improving adaptability while the parametric function design keeps the implementation manageable.
2Productivity
If hypercomplex numbers are used in neural networks, then signal processing efficiency is improved, but the mathematical operations become more complex
Solution Approach 1:
The patent segments the hypercomplex number processing into distinct parametric functions - one for the real component and another for the imaginary component. This segmentation allows the complex mathematical operations to be broken down into manageable, modular functions that can be implemented efficiently in the neural network architecture.
Solution Approach 2:
The patent creates a universal neural network architecture using parametric functions that can handle both real and imaginary components of hypercomplex numbers. This multi-functional approach allows the same network structure to process different types of data (real numbers, complex numbers, quaternions) by simply changing the parametric function configuration, thereby improving signal processing efficiency across multiple domains.
Data Source
AI summary
A hypercomplex operation device according to an embodiment includes processing circuitry. The processing circuitry acquires data including a hypercomplex number, inputs a parametric function in which a function form for a first component and a function form for a second component that is different from the first component differ from each other, inputs the data including a hypercomplex number, to apply to the parametric function, and thereby outputs output data.