loss: Loss and regularisation#

Loss functions and regularisations.

The loss submodule is a collection of implementations of differentiable functions that map from arbitrary inputs to scalar-valued outputs. These scalar outputs can provide a starting point for a backward pass through a differentiable program model. Functionality is provided for various measures of interest to functional brain mapping and other contexts.

Helper wrappers allow packaging of multiple loss objectives in a single call. Each wrapped objective can be selectively applied to a subset of input tensors using LossApply, LossArgument, and LossScheme functionality.