Adaptive k-NN Anomaly Detection With Threshold Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection systems face challenges in accurately distinguishing between good and bad objects due to data imbalance, with many good objects outweighing bad objects, leading to difficulties in constructing a model that effectively identifies anomalies.
Innovation Solution
A method involving a feature extractor neural network and k-nearest neighbors (k-NN) model that adapts a good-vs-bad threshold and updates the k-NN model core set based on human inspection of false alarms, incrementally adjusting the threshold and model to improve anomaly detection performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a k-NN model is trained using only good part data, then the model can efficiently capture characteristics of good objects, but the model struggles to accurately distinguish between good and bad objects due to data imbalance
Solution Approach 1:
The patent implements a feedback mechanism where human inspectors review false alarm classifications and provide feedback to update the training data set. This feedback loop allows the model to learn from its mistakes and adapt to better distinguish between good and bad objects, resolving the contradiction between initial training efficiency and subsequent classification accuracy.
Solution Approach 2:
The patent makes the training data set dynamic by continuously adding newly identified good parts that were previously misclassified as anomalies. This dynamic update process allows the model to adapt to changing data distributions and improve its ability to distinguish between good and bad objects over time, addressing the adaptability issue.
2Reliability
If the anomaly detection threshold is set low to reduce false alarms, then more anomalies are detected, but the number of false anomaly classifications increases
Solution Approach 1:
The patent dynamically adjusts the anomaly detection threshold based on the evolving training data set. As the model learns from feedback and updates its understanding of good objects, the threshold is adapted to maintain optimal sensitivity while reducing false alarms, resolving the contradiction between detection sensitivity and false classification rate.
Solution Approach 2:
The patent changes the threshold parameter adaptively based on model performance and data distribution. By updating the threshold in response to feedback from human inspectors and model improvements, the system maintains high anomaly detection sensitivity while minimizing false anomaly classifications.
3Reliability
If the training data set is continuously updated with new good parts, then the model accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent segments the training data into a core set of good part data that is updated incrementally. By organizing the data structure to allow efficient additions and queries, the system manages computational complexity while maintaining high model accuracy through continuous updates.
Solution Approach 2:
The system performs self-service by automatically incorporating feedback from human inspectors to update its own training data set. This self-updating mechanism improves model accuracy while managing complexity through automated processes that eliminate the need for manual retraining from scratch.
Data Source
AI summary
A method for anomaly detection from object images. A feature extractor provides feature data characterizing part images, including good parts and bad parts. Training data for good parts is used to create a k-nearest neighbors (k-NN) model core set. Adaptation data for some good and some bad parts is evaluated by a k-NN module to determine an anomaly score from the feature data. A good-vs-bad threshold for the anomaly score is initially set to a low value. When an anomaly score over the threshold is detected, a human inspects the corresponding image to determine whether the subject part is good or bad. If the part is good and the anomaly was false, the k-NN model core set is updated to include the feature data for the subject part, and the threshold is adjusted higher. After every adaptation, the k-NN model is applied to a test data set to evaluate performance.