DevNetDetector<T>

Models & Types · AiDotNet.AnomalyDetection.NeuralNetwork

Implements DevNet (Deep Anomaly Detection Network) for end-to-end anomaly scoring.

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

DevNet learns to directly output anomaly scores using a deviation network. It combines feature learning and anomaly scoring in a single network, using reference points and deviation loss for training.

How It Works

The algorithm works by:

  1. Learn feature representations through neural network
  2. Use Gaussian reference points to define normalcy
  3. Train with deviation loss to produce anomaly scores directly

When to use:

  • When you want end-to-end anomaly scoring
  • Tabular data with known anomaly labels (semi-supervised)
  • When reconstruction-based methods don’t work well

Industry Standard Defaults:

  • Hidden dimensions: 64
  • Output dimensions: 1
  • Epochs: 50
  • Learning rate: 0.001
  • Contamination: 0.1 (10%)

Reference: Pang, G., Shen, C., and van den Hengel, A. (2019). “Deep Anomaly Detection with Deviation Networks.” KDD.

Constructors

Constructor Summary
DevNetDetector(int hiddenDim, int epochs, double learningRate, double marginScale, double contamination, int randomSeed) Creates a new DevNet anomaly detector.

Properties

Property Summary
int HiddenDim { get; } Gets the hidden dimensions.

Methods

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