AI Model Training Using Sinusoidal Neuron Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional AI-based model training using back propagation requires significant resources and time due to the need for adjusting weights one-by-one based on error rates, which is inefficient compared to existing methods.
Innovation Solution
The use of a continuous feedforward sinusoid to abstract and adjust the positions of neurons in AI-based models, adjusting weights based on the distance between neurons to correct misclassifications, thereby reducing training time and resource requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If back propagation is used to train AI-based models by adjusting weights one-by-one based on error rates, then the model can be trained to achieve desired accuracy, but the training process requires significant resources and time
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing neuron position adjustments based on sinusoidal signals before the actual training process. The sinusoidal signal is generated in advance with specific frequency and amplitude, and the corresponding weight adjustments are pre-determined, allowing the training to proceed more efficiently without time-consuming iterative calculations during execution.
Solution Approach 2:
The patent changes the parameter approach from traditional back propagation by introducing sinusoidal signal parameters (frequency, amplitude, phase) to control weight adjustments. Instead of adjusting weights based on error rates through iterative gradient descent, the system uses sinusoidal parameter variations to directly modify neuron positions and weights, significantly reducing training time while maintaining accuracy.
2Reliability
If back propagation is used to train AI-based models by adjusting weights one-by-one based on error rates, then the model can be trained to achieve desired accuracy, but significant computational resources are required
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing neuron position adjustments based on sinusoidal signals before the actual training process. The sinusoidal signal is generated in advance with specific frequency and amplitude, and the corresponding weight adjustments are pre-determined, allowing the training to proceed more efficiently without time-consuming iterative calculations during execution.
Solution Approach 2:
The patent replaces the mechanical iterative weight adjustment process of back propagation with a sinusoidal signal-based system. Instead of using gradient descent calculations that require extensive computational iterations, the system substitutes this with sinusoidal signal processing that directly determines weight adjustments, reducing computational resource requirements while maintaining training effectiveness.
3Productivity
If traditional back propagation methods are used for training, then weight adjustment can be performed systematically, but the process is inefficient compared to existing methods
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing neuron position adjustments based on sinusoidal signals before the actual training process. The sinusoidal signal is generated in advance with specific frequency and amplitude, and the corresponding weight adjustments are pre-determined, allowing the training to proceed more efficiently without time-consuming iterative calculations during execution.
Solution Approach 2:
The patent changes the parameter approach from traditional back propagation by introducing sinusoidal signal parameters (frequency, amplitude, phase) to control weight adjustments. Instead of adjusting weights based on error rates through iterative gradient descent, the system uses sinusoidal parameter variations to directly modify neuron positions and weights, significantly reducing training time while maintaining accuracy.
Data Source
AI summary
Methods, apparatus, systems, and articles of manufacture to train an artificial intelligence-based model are disclosed. An example apparatus includes memory; computer readable instructions; and processor circuitry to execute the computer readable instructions to: generate a location value for a neuron in an AI-based model; adjust a characteristic a sinusoidal signal based on a misclassification output by the AI-based model; determine that a trajectory of the sinusoidal signal is within a threshold distance of the location value; adjust the location value in response to the trajectory being within the threshold distance; and adjust a weight that corresponds to the neuron based on the adjusted location value.