Non-Monetary Donation Classification Through Weight Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to efficiently process and classify non-monetary donations during disasters, leading to unpredictable and inefficient supply management.
Innovation Solution
A machine learning model is trained with a reduced number of input parameters and hidden neurons, using weight ranking to optimize classification of non-monetary donations, enabling accurate and resource-efficient processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning model uses all input parameters and neurons for classification, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent extracts and removes redundant input parameters and hidden neurons from the machine learning model through systematic elimination. This is achieved by training the model, analyzing weight magnitudes to identify insignificant parameters, removing those parameters and their associated neurons, and retraining to achieve the same classification accuracy with a reduced model structure.
Solution Approach 2:
The patent changes the parameters of the neural network model by adjusting the number of input parameters and hidden neurons based on weight analysis. The model evolves from a full-parameter configuration to an optimized configuration with fewer parameters, maintaining classification precision while reducing complexity through iterative parameter adjustment.
2Measurement precision
If a machine learning model processes all input parameters, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent extracts and eliminates redundant input parameters through weight analysis, removing parameters with insignificant weights that contribute minimally to classification accuracy. This reduction in parameter count directly decreases processing time while maintaining the same level of measurement precision.
3Measurement precision
If a machine learning model uses more neurons in the hidden layer, then measurement precision is improved, but use of energy increases
Solution Approach 1:
The patent extracts and removes redundant hidden neurons from the model by analyzing weight magnitudes. Neurons with consistently small weights across training iterations are identified as redundant and eliminated, reducing computational energy requirements while preserving classification accuracy.
Solution Approach 2:
The patent optimizes the number of hidden neurons by adjusting this structural parameter based on performance and energy consumption analysis. The model transitions from a high-neuron configuration consuming more energy to an optimized configuration with fewer neurons that achieves the same accuracy with lower energy usage.
4Measurement precision
If extensive parameter verification is performed, then measurement precision is improved, but productivity decreases
Solution Approach 1:
The patent extracts and removes redundant parameters that require verification, identifying them through weight analysis. By eliminating parameters with insignificant weights, the system reduces the verification burden while maintaining classification accuracy, thereby improving processing efficiency and productivity.
Data Source
AI summary
The present disclosure relates to a method for classifying a non-monetary donation. The method comprises: training a machine learning model comprising input neurons arranged in an input layer, each input neuron being operable for inputting an input value (Xn) of a parameter of the donation, and at least one hidden layer comprising multiple hidden neurons, each hidden neuron being operable for calculating a hidden layer value based on at least one input value and weights (w(Xn, Ym)) associated with the respective input neurons. The number of input parameters may be reduced based on ranking the weights (w(Xn, Ym)) related to the individual input parameters. The number of neurons of the hidden layer may be reduced based on ranking the weights related to the individual neurons of the hidden layer. The trained model, the reduced input parameters, and reduced hidden neurons may be provided for enabling the classification using the trained model.