Distance Loss Training for Monotonic Emergency Report Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional loss functions, such as mean square error and cross-entropy, are not suitable for monotonic label classification in emergency report analysis, leading to varying performance of classifiers based on training conditions.
Innovation Solution
A distance loss function is used to train a monotonic multi-label classification model, which calculates a weight based on the distance between target and predicted labels, improving the model's performance by assigning higher weights to distant predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional loss functions (mean square error or cross-entropy) are used for training the classifier, then the model can be trained with standard approaches, but the classification performance for monotonic labels varies and is not optimal
Solution Approach 1:
The patent changes the loss function parameter from conventional options (mean square error, cross-entropy) to a distance-based loss function that calculates the distance between predicted and actual monotonic labels. This parameter change directly addresses the inadequacy of conventional loss functions for monotonic label classification, improving both reliability and measurement precision by incorporating the ordinal nature of urgency levels into the training objective.
2Measurement precision
If a distance loss function is used to improve classification accuracy, then the model better handles monotonic labels, but the training process becomes more computationally complex
Solution Approach 1:
The patent modifies the loss function parameter to use distance-based calculation, which improves measurement precision for monotonic label classification. While this increases training complexity compared to standard loss functions, the complexity is managed through efficient distance calculations that leverage the structured nature of monotonic labels, achieving better accuracy with acceptable computational overhead.
Data Source
AI summary
Provided is a method of training a monotonic multi-label classification model for improving the performance of emergency report analysis. The method includes inputting training data into a monotonic multi-label classification model based on a machine learning model to generate a prediction probability matrix for each of preset monotonic multi-labels, inputting a target value matrix corresponding to the training data and the prediction probability matrix into a predetermined distance loss function to calculate a loss, and training the monotonic multi-label classification model based on the loss.