public interface RegressionLoss<T extends AbstractFloat = Float> extends SupervisedLoss<T, T>
Mean regression objective consuming targets in the model's floating precision. compute(outputs, targets, gradientScale) is inherited from SupervisedLoss: it returns an unscaled mean loss and a mean-reduced output derivative, multiplied by a positive finite gradientScale (one by default).
loss as RegressionLoss<Float> = MeanSquaredError<Float>()