Achlioptas fills the input matrix with values according to the mthod described on "Database-friendly random projections: Johnson-Lindenstrauss with binary coins", by Dimitris Achlioptas 2001 (https://core.ac.uk/download/pdf/82724427.pdf) The matrix is returned for convenience.

Constant fills the input matrix with the value n.

Gain returns a coefficient that help to initialize the params in a way to keep gradients stable.

Normal fills the input matrix with random samples from a normal (Gaussian) distribution.

Ones fills the input matrix with the scalar value `1`.

Uniform fills the input matrix m with a uniform distribution where a is the lower bound and b is the upper bound.

XavierNormal fills the input matrix with values according to the method described in "Understanding the difficulty of training deep feedforward neural networks" - Glorot, X.

XavierUniform fills the input `m` with values according to the method described in `Understanding the difficulty of training deep feedforward neural networks` - Glorot, X.

Zeros fills the input matrix with the scalar value `0`.