BetaNLL

class torch_uncertainty.losses.BetaNLL(beta=0.5, reduction='mean')

The Beta Negative Log-likelihood loss.

Parameters:

• beta (float) – TParameter from range [0, 1] controlling relative

• points (weighting between data) –

• to (where 0 corresponds) –

• weighting. (high weight on low error points and 1 to an equal) –

• reduction (str, optional) – specifies the reduction to apply to the

• output – 'none' | 'mean' | 'sum'.

Reference:

Seitzer, M., Tavakoli, A., Antic, D., & Martius, G. (2022). On the pitfalls of heteroscedastic uncertainty estimation with probabilistic neural networks. https://arxiv.org/abs/2203.09168.