Loss Functions

Describe the main training losses, the assumptions behind them, their reductions and masks, and their gradients.