VAEDetector<T>

Models & Types · AiDotNet.AnomalyDetection.NeuralNetwork

Detects anomalies using Variational Autoencoder (VAE).

Inheritance: ModelBase<T, Matrix<T>, Vector<T>> → AnomalyDetectorBase<T> → VAEDetector<T>

Implements: IGeneratedParameterRegistrar<T>

For Beginners

A VAE is a generative neural network that learns to encode data into a lower-dimensional probabilistic latent space and decode it back. Anomalies are points that are poorly reconstructed or fall in low-probability regions of the latent space.

How It Works

The algorithm works by:

  1. Train encoder to map data to latent distribution (mean + variance)
  2. Train decoder to reconstruct from latent samples
  3. Score combines reconstruction error and KL divergence

When to use:

  • Complex, high-dimensional data
  • When you want probabilistic anomaly scores
  • Image, text, or structured data anomalies

Industry Standard Defaults:

  • Latent dimensions: 10
  • Hidden dimensions: 64
  • Learning rate: 0.001
  • Epochs: 100
  • Contamination: 0.1 (10%)

Reference: Kingma, D.P., Welling, M. (2014). “Auto-Encoding Variational Bayes.” ICLR. An, J., Cho, S. (2015). “Variational Autoencoder based Anomaly Detection.”

Constructors

Constructor Summary
VAEDetector(int latentDim, int hiddenDim, int epochs, double learningRate, double contamination, int randomSeed) Creates a new VAE anomaly detector.

Properties

Property Summary
int HiddenDim { get; } Gets the hidden layer dimensions.
int LatentDim { get; } Gets the latent space dimensions.

Methods

Method Summary
Vector<T> ScoreAnomalies(Matrix<T> X)
void Fit(Matrix<T> X)