MSE

Loss variant computing the mean squared error between the model output and the targets. It penalizes large errors quadratically and is a standard objective for regression. Compatible with the regression policy.

Configuration

Selected as name: mse.

Option	Description	Type	Default
reduction	How per-sample losses are reduced.	mean | sum	mean

components:
  loss:
    name: mse
    kwargs: {}