Boolean Neural Network Architecture for Low-Complexity Logic Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural networks rely on real-number inputs and outputs, which can be computationally intensive and may not efficiently process Boolean data, limiting their applicability in certain domains.
Innovation Solution
A neural network architecture utilizing Boolean computational nodes that receive and generate Boolean inputs and outputs, applying Boolean functions and weighting functions to process data efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional neural networks use real-number inputs and outputs, then they can perform complex computational operations, but they become computationally intensive and inefficient when processing Boolean data
Solution Approach 1:
The patent changes the data type parameter from real numbers to Boolean values throughout the neural network system. This includes Boolean inputs, Boolean weights, Boolean activation functions, and Boolean outputs. By changing this fundamental parameter, the system achieves efficient Boolean data processing while maintaining neural network functionality through logical operations instead of arithmetic computations.
2Adaptability or versatility
If conventional neural networks process Boolean data using real-number operations, then they can leverage existing computational frameworks, but they fail to efficiently process Boolean data and are limited in applicability
Solution Approach 1:
The patent substitutes arithmetic mechanical operations with logical operations. Instead of using real-number addition, multiplication, and activation functions, the system employs Boolean logical operations (AND, OR, NOT, XOR) to perform computations. This substitution enables the neural network to natively process Boolean data efficiently while maintaining the structural and functional principles of neural networks.
3Reliability
If conventional neural networks are trained using backpropagation with gradient descent, then they can learn from training data, but the training process becomes computationally intensive
Solution Approach 1:
The patent inverts the traditional training approach by using forward propagation to compute outputs and then directly comparing Boolean outputs to desired outputs to determine weight updates. Instead of computing gradients through backpropagation, the system uses Boolean difference logic to identify which weights need to be flipped (0 to 1 or 1 to 0) to correct errors, significantly reducing computational resources required for training.
Data Source
AI summary
First Boolean computational nodes in a first layer of a neural network receive first Boolean inputs, and the first Boolean computational nodes generate first Boolean outputs based on the plurality of first Boolean inputs. Second Boolean computational nodes in a second layer of the neural network receive second Boolean inputs. The second Boolean inputs are based on the first Boolean outputs generated by the first Boolean computational nodes. The second Boolean computational nodes generate weighted second Boolean inputs by respectively applying respective Boolean weighting functions to at least some of second Boolean inputs, and the second Boolean computational nodes generate second Boolean outputs based on the weighted second Boolean inputs.