Enum UncertaintyQuantificationMethod

Namespace

AiDotNet.Enums

Assembly

AiDotNet.dll

Defines the supported uncertainty quantification strategies for inference.

public enum UncertaintyQuantificationMethod

Fields

Auto = 0

Automatically selects a suitable method when possible, otherwise falls back to deterministic predictions.

BayesianNeuralNetwork = 3

Uses Bayesian neural network sampling (e.g., Bayes-by-Backprop style layers) to estimate uncertainty.

ConformalPrediction = 6

Uses conformal prediction to produce statistically valid intervals (regression) or prediction sets (classification).

DeepEnsemble = 2

Uses a deep ensemble (multiple independently trained models) to estimate uncertainty.

LaplaceApproximation = 4

Uses a Laplace approximation (typically diagonal) over model parameters to sample predictions.

MonteCarloDropout = 1

Uses Monte Carlo Dropout by enabling dropout at inference and sampling multiple forward passes.

Swag = 5

Uses SWAG (Stochastic Weight Averaging-Gaussian) to sample model parameters and estimate uncertainty.

Remarks

These options control how the system estimates predictive uncertainty during inference.

For Beginners: Some models can also tell you how confident they are. This enum lets you choose the strategy used to estimate that confidence.