RbfWidthInitialization

Enums · AiDotNet.Enums

How RBFLayer sets each unit’s initial width.

For Beginners

Each RBF unit responds to inputs near its centre, and its width sets how near. A width that is far too small makes the unit respond to almost nothing, so its output is zero for every input and it never trains. These choices decide how the widths start out before training adjusts them.

Fields

Field Summary
RbfWidthInitialization CenterSpread Every width is d_max / sqrt(2M), where d_max is the largest distance between two initial centres and M is the number of centres (Broomhead and Lowe 1988; Haykin, Neural Networks, section 5.10).
RbfWidthInitialization Uniform Each width is drawn from Uniform(0, 1), the layer’s original behaviour.
int value__