Glorot Normal Initialiser

glorot_normal_init_type

glorot_normal_init_type()

Also known as Xavier normal initialisation. Draws samples from a normal distribution centered at zero.

\[W \sim \mathcal{N}\left(0, \sqrt{\frac{2}{n_{in} + n_{out}}}\right)\]

where \(n_{in}\) is the number of input units and \(n_{out}\) is the number of output units.

This initialisation helps maintain the variance of activations across layers and is well-suited for networks using sigmoid or tanh activations.

Shape:

Initialises weights based on the shape provided during layer setup.