Statistical Tutorials

Pipelines that learn from background statistics alone — no gradient steps, no optimizer. Statistical nodes accumulate moments (mean, covariance, histograms) during initialization and use them at inference time. Fast to train, interpretable, and strong baselines.

In this section

• RX Anomaly Detection — classical Mahalanobis-distance anomaly detector. The canonical statistical baseline for hyperspectral anomaly detection.
• Channel Selection — two-phase training that learns which wavelengths matter. Statistical warm-up sets the initial weights; gradient refinement (later phase) sharpens them.
• Blood Perfusion — normalised-difference (NDVI-style) two-band differential for tissue visualisation.

See also

• Concepts → Training — the two-phase training model.
• Gradient tutorials — the other half of the training story.