Anomaly Detection Model Multi-Task Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection models are limited to performing a single task and require multiple models to be trained for different tasks, leading to increased development costs and complexity when tasks share overlapping features.
Innovation Solution
A training method for an anomaly detection model using a generative and discriminative model that calculates network loss results based on original images and task information, allowing a single model to perform multiple tasks by adjusting parameters according to weighted sums of network losses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single anomaly detection model is used to perform multiple anomaly detection tasks, then the development cost and complexity are reduced, but the model can only effectively perform tasks when anomaly features do not significantly overlap with normal features of other tasks
Solution Approach 1:
The patent implements multi-functionality by designing a single anomaly detection model that can perform multiple detection tasks through task information embedding. The model processes different tasks by incorporating task-specific information into the input data, allowing one model structure to handle multiple functions without requiring separate models for each task, thus reducing overall system complexity while maintaining versatility
2Reliability
If multiple anomaly detection models are established for different tasks, then each task can be optimized independently, but the development cost and complexity are dramatically increased
Solution Approach 1:
The patent applies merging by combining multiple task-specific detection capabilities into a single unified model structure. Instead of training separate models for each task, the approach merges all task requirements into one model that processes task information alongside image data, reducing the number of models from multiple to one while maintaining the ability to perform each task with appropriate accuracy
3Measurement precision
If anomaly features of one task overlap with normal features of another task, then a single model cannot perform both tasks effectively, but training separate models increases cost and complexity
Solution Approach 1:
The patent introduces task information as an intermediary element that mediates between overlapping features of different tasks. By embedding task-specific information into the input data, the model can distinguish between features that would otherwise be ambiguous or overlapping, enabling accurate discrimination without requiring separate models for each task
Data Source
AI summary
A training method for an anomaly detection model and an electronic device using the same are provided. The anomaly detection model includes a generative model and a discriminative model. The training method for the anomaly detection model includes the following steps. One of a plurality of original images and one of a plurality of task information are used as a training sample. The training sample is input to the generative model and the discriminative model to calculate a plurality of network loss results corresponding to the training sample. If the original image of the training sample does not match the task information of the training sample, a first loss function is obtained based on a weighted sum of reciprocals of the network loss results, and the generative model is trained according to the first loss function.