Neural Network Activation Coefficient Optimization for Synthetic Resin Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing prediction technologies for chemical reactions, such as those involving synthetic resins, lack effective design methods and optimization techniques for neural network models, leading to suboptimal prediction accuracy.
Innovation Solution
A method for prediction using an information processing device that trains a neural network model with an input layer, an intermediate layer, and an output layer, where the coefficient of the activation function of the intermediate layer is larger than that of the output layer, optimizing the learning process and improving prediction accuracy for chemical reactions of synthetic resins.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional neural network models are used for prediction without specific design methods, then the model can be implemented with standard configurations, but the prediction accuracy for chemical reactions of synthetic resins is suboptimal
Solution Approach 1:
The patent applies parameter changes by optimizing specific neural network parameters including setting the number of hidden layers to 1-3, configuring hidden layer nodes to be 1.1-6 times the number of explanatory factors, and setting activation function coefficients α1>α2. These parameter optimizations directly improve prediction accuracy for chemical reaction outcomes while maintaining manageable model complexity through systematic parameter selection rather than trial-and-error approaches.
Solution Approach 2:
The patent implements local quality by differentiating the activation function coefficients across different layers (α1 for hidden layers, α2 for output layer where α1>α2). This local differentiation allows each layer to perform its specific function optimally - hidden layers benefit from higher coefficients for feature extraction while the output layer uses lower coefficients for precise prediction, thereby improving overall prediction accuracy without uniformly increasing model complexity.
2Measurement precision
If the neural network model uses uniform activation function coefficients across all layers, then the model structure is simpler, but the learning process is not optimized leading to suboptimal prediction results
Solution Approach 1:
The patent changes the parameter configuration by introducing different activation function coefficients for different layers (α1 for hidden layers, α2 for output layer). This parameter differentiation optimizes the learning process by allowing hidden layers to learn complex patterns more effectively while the output layer provides stable predictions, thereby improving prediction accuracy for chemical reactions despite the increased configuration complexity.
Solution Approach 2:
The patent applies dynamics by making the activation function coefficients adaptable to different layer functions. The coefficient α1 for hidden layers is set higher to facilitate robust feature extraction and pattern recognition, while α2 for the output layer is set lower to ensure stable and accurate final predictions. This dynamic parameter assignment optimizes the learning process across different stages of the network.
3Measurement precision
If the number of hidden layer nodes is not optimized, then the model configuration is simpler, but the prediction capability for chemical reactions is insufficient
Solution Approach 1:
The patent optimizes the number of hidden layer nodes by setting it to be 1.1 to 6 times the number of explanatory factors. This parameter optimization ensures sufficient model capacity to capture complex chemical reaction patterns while avoiding excessive complexity. The systematic relationship between input dimensions and hidden layer size provides a principled approach to model design that balances prediction accuracy with model simplicity.
Data Source
AI summary
A method for prediction executed by an information processing device includes: training a neural network model based on actual data on a prediction target; and predicting an objective factor related to the prediction target based on explanatory factors related to the prediction target by the neural network model. The neural network model includes an input layer, an intermediate layer, and an output layer, and a coefficient of an activation function of the intermediate layer is larger than a coefficient of an activation function of the output layer.