qumphy.models.utils.kgloss module

class qumphy.models.utils.kgloss.KGLoss(*args: Any, **kwargs: Any)[source]

Bases: Module

Equation 12 from https://arxiv.org/abs/1703.04977 Models aleatoric uncertainty for classification tasks

forward(logits, noise, target)[source]

Compute the loss for aleatoric uncertainty.

Parameters:

logits (torch.Tensor) – The output of the model before softmax
noise (torch.Tensor) – The output of the model for the aleatoric uncertainty
target (torch.Tensor) – The target of the model

Returns:

loss – The loss of the model

Return type:

torch.Tensor

class qumphy.models.utils.kgloss.KGLoss_unified_prediction(*args: Any, **kwargs: Any)[source]