Anomaly Detection Model Parameter Adjustment via Feedback Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for anomaly detection face issues with maintaining consistency and continuity of anomaly outputs before and after model updates, leading to suboptimal performance when updated with new anomalous data during operation.
Innovation Solution
A learning apparatus that calculates an anomaly degree using a loss function with an adjustment parameter based on previous loss values, balancing the normal and anomaly label terms to ensure consistent model updates and improve detection performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the model is updated each time with new anomalous data during operation, then the detection performance is improved, but the consistency of the model before and after update is lost
Solution Approach 1:
The patent implements feedback by calculating a loss value based on previous loss values and using this feedback to adjust the adjustment parameter. The loss function incorporates historical loss information (L_prev) to provide feedback on model performance trends, enabling the system to adapt the learning rate dynamically while maintaining consistency across updates.
Solution Approach 2:
The adjustment parameter is made dynamic by adjusting it based on the calculated loss value and previous loss values. This dynamic adjustment allows the model to adapt its learning behavior during operation, improving detection performance while maintaining consistency through controlled adaptation rather than rigid fixed parameters.
2Measurement precision
If the adjustment parameter is increased to balance anomaly label term, then the anomaly detection capability is improved, but the loss value becomes unstable
Solution Approach 1:
The system uses feedback from previous loss values (L_prev) to dynamically adjust the adjustment parameter. This feedback mechanism ensures that the parameter adapts based on actual model performance trends, preventing instability by using historical performance information to guide parameter adjustments.
Solution Approach 2:
The adjustment parameter is changed dynamically based on the calculated loss value and previous loss values. This parameter change strategy allows the system to optimize anomaly detection capability while maintaining loss stability by adapting the parameter to current model performance rather than using fixed aggressive values.
3Adaptability or versatility
If the model is updated frequently with operation data, then the adaptability to new anomalies is improved, but the continuity of anomaly degrees is lost
Solution Approach 1:
The loss function incorporates previous loss values to provide continuous feedback across model updates. This feedback ensures that anomaly degree measurements remain continuous and comparable across updates, as the adjustment parameter evolves smoothly based on historical performance rather than making abrupt changes.
Solution Approach 2:
The patent maintains continuity of useful action by ensuring that the adjustment parameter changes continuously based on loss value trends. This continuous adaptation preserves the meaningfulness and comparability of anomaly degrees across multiple updates while maintaining adaptability to new anomaly patterns.
Data Source
AI summary
According to one embodiment, a learning apparatus includes a processor. The processor acquires data with a label indicating whether the data is normal data or anomalous data. The processor calculates an anomaly degree indicating a degree to which the data is the anomalous data using an output of a model for the data. The processor calculates a loss value related to the anomaly degree using a loss function based on an adjustment parameter based on a previously calculated loss value and the label. The processor updates a parameter of the model so as to minimize the loss value.