Variational Autoencoder Training for Distributed Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data generation techniques for training variational autoencoders (VAEs) are inefficient and fail to maintain the underlying statistical properties of original datasets, particularly in sensitive or complex systems like antenna networks, leading to missed anomalies and impractical validation testing.
Innovation Solution
A method for training VAEs using game theoretic optimization and probabilistic models to generate synthetic data that retains the statistical properties of the original dataset, utilizing a latent space with a 3-D point scatter plot and solvers like Adam, SGDM, and RMSProp to update weights and biases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If random oversampling is used to augment training datasets, then the dataset balance is improved, but the statistical meaningful representation of the original dataset is lost
Solution Approach 1:
The patent uses Variational Autoencoders to generate synthetic copies of minority class samples that preserve the underlying statistical distribution. Instead of simple duplication, the VAE learns the probability distribution of the data and generates new samples that maintain statistical fidelity, thus balancing the dataset without losing meaningful representation
Solution Approach 2:
The patent modifies the data generation process by changing from random copying to probabilistic generation using VAEs. The model learns parameters of the data distribution and generates samples with controlled statistical properties, transforming the oversampling process from naive duplication to statistically sound synthesis
2Reliability
If current data is compared with known normal data for thousands of antennae, then anomaly detection capability is improved, but the time required for inspection increases to months
Solution Approach 1:
The patent applies VAEs to pre-process and compress normal antenna data into a latent space representation before actual anomaly detection. This preliminary encoding of normal patterns enables rapid comparison against new data, transforming a months-long inspection process into a much faster operation while maintaining detection accuracy
Solution Approach 2:
The patent extracts the essential normal patterns from large volumes of antenna data and stores them in a compressed latent space representation. This extraction of core normal behavior enables efficient anomaly detection by comparing against the extracted patterns rather than processing all original data, significantly reducing inspection time
3Object-affected harmful factors
If synthetic data is generated to protect sensitive data, then data security is improved, but the ability to retain underlying statistical properties is compromised
Solution Approach 1:
The patent uses VAEs to generate synthetic copies of sensitive data that preserve statistical properties. The model learns the probability distribution of the original data and generates new samples that maintain the same statistical characteristics while containing no actual sensitive information, thus achieving both security and statistical fidelity
4Ease of manufacture
If simple random duplication is used for data generation, then implementation simplicity is improved, but the quality and statistical soundness of training datasets deteriorates
Solution Approach 1:
The patent replaces the simple mechanical copying process with a learned probabilistic model. Instead of direct duplication, the VAE learns the underlying data distribution and generates samples through probabilistic sampling, substituting a complex but effective mechanism for a simple but inadequate one
Data Source
AI summary
Methods of training a variational autoencoder (VAE) to recognize anomalous data in a distributed system are provided. Input image data representative of devices/processes in a distributed system are provided to an encoder of a VAE on a processor. The input image data is compressed, via the processor, using a first plurality of weights with the encoder. A normal distribution of the compressed image data is created in a latent space of the VAE. The compressed image data from the latent space is decompressed using a second plurality of weights with a decoder of the VAE. The decompressed image data from the decoder is optimized. At least the first and second plurality of weights are updated, via the processor, based on the loss detected in the optimized decompressed image data. The above steps are iterated until the decompressed image data possesses substantially the same statistical properties as the input image data.