Machine Learning Anomaly Detection with Reduced Feature Memory Banks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning-based anomaly detection techniques face challenges in effectively learning anomaly data due to the scarcity and uneven distribution of anomaly data, particularly in industrial applications, leading to inefficiencies and reduced accuracy.
Innovation Solution
A machine learning-based anomaly detection method that utilizes normal data for training, employing a reduced feature extraction module and a memory bank to detect anomalies through data amplification and supervised learning, leveraging a feature extraction network and dimensionality reduction network to generate and classify reduced feature data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning classification models are trained with a large amount of normal and anomaly data, then detection accuracy can be improved, but anomaly data exhaustion occurs because anomaly data is difficult to obtain
Solution Approach 1:
The patent creates synthetic anomaly data by copying and transforming normal data through data augmentation techniques. The system generates augmented normal data and artificially creates anomaly data by introducing perturbations to normal data samples, effectively copying the structure of normal data while introducing anomaly characteristics.
Solution Approach 2:
The patent changes data parameters by applying various transformations to normal data to create anomaly samples. This includes modifying data characteristics such as adding noise, changing intensity values, and applying geometric transformations to convert normal data into anomaly data with different parameter configurations.
2Ease of manufacture
If conventional data augmentation and self-supervised learning are used to detect anomaly data, then training with only normal data becomes possible, but detection accuracy is reduced due to insufficient anomaly data representation
Solution Approach 1:
The patent performs preliminary actions by pre-processing normal data through augmentation and creating a comprehensive set of transformed normal data samples before anomaly detection. The system prepares augmented normal data and synthetic anomaly data in advance, establishing a robust training foundation that enables accurate anomaly detection without requiring actual anomaly data collection.
Solution Approach 2:
The patent combines multiple types of data representations and processing techniques to create a composite training framework. It integrates augmented normal data, synthetic anomaly data, and various feature representations into a unified training system that leverages the strengths of different data types and processing methods to achieve high detection accuracy.
3Measurement precision
If feature extraction networks process large amounts of data to generate comprehensive feature representations, then detection accuracy improves, but computational complexity and processing time increase
Solution Approach 1:
The patent extracts only the most relevant and discriminative features from the data using specialized feature extraction networks. Instead of processing all data features equally, the system identifies and extracts key anomaly-related features through learned representations, reducing computational complexity while maintaining detection accuracy by focusing on the most informative features.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A computing device for performing anomaly detection according to the present disclosure includes a memory including at least one memory bank that is a logical area, and a processor configured to train a neural network by using at least one function module, generating a reduced feature extraction module, generating the memory bank on the basis of reduced feature data generated from the reduced feature extraction module, and determining whether inspection data is normal by using the generated reduced feature extraction module and the memory bank.