Expanding Weight Initialization Module

[LazaroHurtado]

Using the proper weight initialization can significantly boost performance gains by accelerating convergence of SGD. With this being said, we should expand our WeightInit enum by incorporating the He and LeCun schemes. Currently, we only offer Glorot/Xavier but this is suboptimal for the ReLu activation function so some tiny changes must be made.

Anyone can claim implementing these features:

• He Weight Initialization
• LeCun Weight Initialization

Here are some extra resources you might find helpful:
Glorot and He discussion
wandb Intro to Weight Inits

[jrego43]

I'll try my hand at implementing He Initialization tomorrow.

[Demba3]

I will be implementing LeCun weight initialization. Please let me know if this causes any conflicts.