Neural Network Anomaly Detection via Score Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of anomaly detection using neural networks is compromised due to the limitations in approximating real data distributions with Gaussian distributions, leading to performance degradation in anomaly detection methods.
Innovation Solution
The method involves generating normal and abnormal noise specific to the data type, calculating scores, and adjusting neural network parameters to minimize differences between these scores, thereby improving the accuracy of anomaly detection by enhancing the approximation of data distributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If Gaussian distribution approximation is used for anomaly detection, then the detection process is simplified, but the detection accuracy deteriorates
Solution Approach 1:
The patent changes the fundamental parameter of data distribution approximation from Gaussian distribution to a more flexible distribution model that can adapt to the actual characteristics of the input data. This is achieved by using learned parameters from the neural network to define the data distribution, allowing the model to capture complex patterns while maintaining computational tractability through the proposed objective function.
2Adaptability or versatility
If score matching is used for learning, then applicability to various models is improved, but learning difficulty increases
Solution Approach 1:
The patent introduces denoising autoencoders as an intermediary mechanism to simplify the score matching learning process. By framing the learning objective as denoising reconstruction, the complex score matching problem becomes more tractable while maintaining the ability to learn various types of models. The autoencoder structure provides a practical framework for implementing score matching without directly solving the difficult optimization problems.
Data Source
AI summary
An anomaly detection method includes generating second data by adding normal noise to first data; generating third data by adding abnormal noise to the first data; inputting the first data, the second data, and the third data to a neural network; calculating a first normal score, a second normal score, and a third normal score; calculating a third difference based on a first difference and a second difference, the first difference being based on a difference between the first normal score and the second normal score, the second difference being based on a difference between the first normal score and the third normal score; changing the neural network so that the third difference becomes smallest; inputting, to the changed neural network, fourth data that is unknown in terms of whether the fourth data is normal or abnormal; and determining whether the fourth data is normal or abnormal.