Guided Neural Network Training via Convolutional Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional guided training of deep neural network models often results in a second feature domain dominating the model output, undermining the goal of improving the model based on multiple domains of feature data.
Innovation Solution
The method involves receiving supplementary domain feature data and fusing it with feature maps from a convolution layer to form fused feature maps, which are then provided to a subsequent convolution layer, thereby balancing the impact of supplementary features and guiding the convolution layers effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If supplementary domain feature data is provided to a fully connected layer in conventional guided training, then the model can incorporate additional domain knowledge, but the supplementary domain features become a dominant driver of the model output, undermining the goal of improving the model based on multiple domains of feature data
Solution Approach 1:
The patent introduces convolution layers as intermediary components between the supplementary domain features and the fully connected layer. These convolution layers process and transform the supplementary features before they reach the decision-making layers, preventing direct dominance while maintaining their informative value. The convolution operations act as a buffer that integrates multiple domains more harmoniously.
Solution Approach 2:
The patent segments the feature processing pipeline by separating the handling of primary domain features and supplementary domain features into different convolutional pathways. This segmentation allows each domain to be processed independently through dedicated convolution layers before being combined, preventing any single domain from dominating the final output.
2Quantity of substance
If supplementary domain features are directly fed to fully connected layers, then the training process can use additional domain information, but the model output becomes overly influenced by the supplementary domain, reducing overall model performance
Solution Approach 1:
The patent applies preliminary processing to supplementary domain features by passing them through convolution layers before they are combined with primary domain features. This preliminary action transforms the raw supplementary features into a more suitable representation that can be effectively integrated without overwhelming the model output, thereby maintaining precision while utilizing abundant data.
Data Source
AI summary
Aspects described herein provide a method of performing guided training of a neural network model, including: receiving supplementary domain feature data; providing the supplementary domain feature data to a fully connected layer of a neural network model; receiving from the fully connected layer supplementary domain feature scaling data; providing the supplementary domain feature scaling data to an activation function; receiving from the activation function supplementary domain feature weight data; receiving a set of feature maps from a first convolution layer of the neural network model; fusing the supplementary domain feature weight data with the set of feature maps to form fused feature maps; and providing the fused feature maps to a second convolution layer of the neural network model.