VAE Anomaly Detection via Latent Distribution Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current variational autoencoder (VAE) methods for anomaly detection lack accuracy due to the assumption of a single-variable-independent normal distribution, which does not guarantee that the latent space probability distribution reflects the real space distribution, leading to low anomaly detection accuracy, especially when the real space distribution has a complex shape.
Innovation Solution
An information processing apparatus that corrects the prior probability of the latent variable using the posterior distribution parameter from the encoder, converting the latent variable's probability distribution into a second probability distribution based on the decoder's output error, and outputs this as an estimated value of the input data's probability distribution to improve anomaly detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a single-variable-independent normal distribution is assumed for the latent variable, then the model training is simplified, but the anomaly detection accuracy deteriorates because the latent space probability distribution does not accurately reflect the real space distribution
Solution Approach 1:
The patent changes the parameter representation of the probability distribution from a simple normal distribution to a distribution characterized by cumulative distribution function parameters. Specifically, it uses the relationship between the cumulative distribution function F(x) and its derivative f(x) to transform the probability distribution parameters, allowing the latent space distribution to more accurately reflect the real space distribution while maintaining computational tractability through the use of standard normal distribution transformations.
2Measurement precision
If the latent space probability distribution is transformed to accurately reflect real space distribution, then anomaly detection accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent introduces an intermediary transformation process that uses the cumulative distribution function F(x) and its derivative f(x) as mediators between the latent space and real space probability distributions. This intermediary approach allows for accurate distribution transformation without requiring complex direct mapping, as the transformation can be performed through standard normal distribution functions and their derivatives, which are computationally efficient and well-established in probability theory.
Data Source
AI summary
A non-transitory computer-readable storage medium storing an estimation program that causes at least one computer to execute a process, the process includes inputting an input data into a trained variational autoencoder that includes an encoder and a decoder; converting, into a first probability distribution, a probability distribution of a latent variable that is generated by the trained variational autoencoder according to the input based on a magnitude of a standard deviation output from the encoder; converting the first probability distribution into a second probability distribution based on an output error of the decoder regarding the input data; and outputting the second probability distribution as an estimated value of a probability distribution of the input data.