Neural Network Training With Missing-Tag Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep neural network models suffer from reduced prediction accuracy due to missing data in the training dataset, as current methods like deleting or interpolating with mean/median values fail to consider the complexity and diversity of missing data, leading to loss of authenticity and information.
Innovation Solution
A model training method that retains sample data with missing tags by calculating first and second weight values based on association information, tag missing ratios, and information entropy and mutual information to improve the neural network's prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sample data with missing tags is deleted or interpolated using mean/median values, then the training process can be completed, but the prediction accuracy deteriorates due to loss of authenticity and information
Solution Approach 1:
The patent changes the parameter of weight values assigned to different dimensions based on their missing ratios. By calculating weight values according to the proportion of missing tags in each dimension and using these weights in a weighted loss function, the model adapts to varying data completeness without deleting or interpolating samples, thus maintaining both training reliability and prediction accuracy.
Solution Approach 2:
The patent introduces weight values as an intermediary mechanism between the missing data problem and the training process. These weights act as mediators that adjust the contribution of different dimensions based on their missing ratios, allowing the model to handle incomplete data without requiring deletion or interpolation, thereby preserving authenticity and information.
2Device complexity
If simple deletion or mean/median interpolation is used to handle missing data, then the processing complexity is reduced, but the diversity and complexity of missing data patterns are not fully considered
Solution Approach 1:
The patent dynamically adjusts weight values based on the specific missing ratio patterns of different dimensions. Instead of using a fixed approach like deletion or simple interpolation, the model calculates weight values according to the actual missing data patterns, enabling it to adapt to diverse and complex missing data scenarios while maintaining manageable processing complexity through the weighted loss function.
Data Source
AI summary
A model training method includes: obtaining sample data; determining a tag missing ratio of the training sample in different dimensions and a first weight value of the training sample in different dimensions according to association information between a training sample and a true value tag; determining a second weight value of the training sample according to the tag missing ratio; and inputting the sample data, the first weight value, and the second weight value into a preset neural network model for training, until a loss value of a target loss function of the preset neural network model meets a model convergence condition to obtain a target prediction model.