Losses

MappingLoss	Composite loss implementing the paper's multi-component mapping loss.
BaseLoss	Contract for individual loss components.
LossOutput	Structured output from a loss computation.
TrainingContext	Carry all information loss components need in one object.
AlignmentLoss	Cosine distance between latent vectors and mapper weight summaries.
SmoothnessLoss	Penalize the Frobenius norm of the mapper Jacobian.
StabilityLoss	Penalize prediction sensitivity to small latent perturbations.
TaskLoss	Wrap an arbitrary (predictions, targets) -> scalar callable.
ClassificationLoss	Cross-entropy task loss for classification targets.
RegressionLoss	Mean squared error task loss for regression targets.