Fast Ica Contrast Function Enumeration
Enumerates the contrast functions used in ICA.
Definition
Namespace: Numerics.NET.Statistics.Multivariate
Assembly: Numerics.NET (in Numerics.NET.dll) Version: 9.1.5
C#
Assembly: Numerics.NET (in Numerics.NET.dll) Version: 9.1.5
public enum FastIcaContrastFunction
Remarks
The contrast function determines the nonlinearity applied during the fixed-point iteration in the FastICA algorithm. Different choices affect robustness and convergence speed.
- LogCosh: Uses the hyperbolic tangent function, g(u) = tanh(αu). Suitable for most applications and robust to outliers.
- Exponential: Uses g(u) = u * exp(-u² / 2). More sensitive to outliers, can be useful for super-Gaussian sources.
- Cubic: Uses g(u) = u³. Simple and fast, but less robust to outliers.
Members
LogCosh | 0 |
LogCosh contrast function: g(u) = tanh(αu).
The default and recommended choice for most ICA problems. The parameter α controls the steepness of the nonlinearity. |
Exponential | 1 |
Exponential contrast function: g(u) = u * exp(-u² / 2).
Useful for separating super-Gaussian sources. More sensitive to outliers. |
Cubic | 2 |
Cubic contrast function: g(u) = u³.
Fast to compute, but less robust to outliers. Suitable for sub-Gaussian sources. |