He Normal Initialiser

he_normal_init_type

he_normal_init_type()

Draws samples from a normal distribution designed for layers with ReLU activation.

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

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

This initialisation is specifically designed for networks using ReLU activations and helps prevent vanishing/exploding gradients.

Shape:

Initialises weights based on the shape provided during layer setup.