Deep Neural Network Model Uncertainty-Based Active Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing indicator detection methods face challenges in training models with excellent performance due to high label costs and accuracy issues when dealing with large-scale indicators, as they require extensive labeling, which is costly and difficult to implement.
Innovation Solution
A method that utilizes a deep neural network model to determine the uncertainty of detection results, selects reference indicator data based on this uncertainty, and performs active learning using these data points to optimize the model, thereby reducing the need for extensive labeling and improving performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large-scale indicator data is labeled for model training, then model detection accuracy is improved, but label cost and implementation difficulty increase significantly
Solution Approach 1:
The patent applies partial action by selecting only a subset of indicator data (reference indicator data) for labeling and model training, rather than labeling all available data. The model uses uncertainty metrics to identify and focus on the most valuable samples for labeling, achieving good detection accuracy with reduced label cost and implementation effort
Solution Approach 2:
The model performs self-service by automatically computing uncertainty metrics for each indicator data point and autonomously selecting which samples require labeling. This self-directed sample selection process eliminates the need for manual assessment of all data points, reducing implementation difficulty and label cost while maintaining detection accuracy
2Quantity of substance
If only small-scale indicator data is labeled for model training, then label cost is reduced, but detection accuracy of all indicators cannot be guaranteed
Solution Approach 1:
The patent changes the parameter of sample selection from random or uniform sampling to uncertainty-based sampling. By computing uncertainty metrics and selecting samples with highest uncertainty values as reference indicator data, the model ensures that the limited labeled data provides maximum information gain, maintaining detection accuracy while reducing label cost
Solution Approach 2:
The model implements feedback by using uncertainty metrics to guide the selection of reference indicator data for labeling. The uncertainty computation provides feedback on which samples are most informative, allowing the model to iteratively improve detection accuracy by focusing labeling efforts on the most critical samples rather than distributing labels uniformly across all data
Data Source
AI summary
Embodiments of this application disclose a model training method and related apparatuses in the field of artificial intelligence. The method includes: acquiring at least one to-be-detected indicator data in target business scenarios; specific to each to-be-detected indicator data, determining, by a deep neural network model, uncertainty of a detection result corresponding to the to-be-detected indicator data, where the uncertainty is used for representing the degree of reliability of the detection results, and the detection results are determined by the deep neural network model according to the to-be-detected indicator data; selecting reference indicator data from the at least one to-be-detected indicator data according to the uncertainty of the detection result corresponding to each of the at least one to-be-detected indicator data, and acquiring label detection results corresponding to the reference indicator data; and training the deep neural network model based on the reference indicator data and their corresponding label detection results, so as to obtain a target indicator detection model applicable to the target business scenarios. The method can reduce the cost for training the indicator detection model.