Losses: Batch Inverse Variance (BIV)¶
-
class
losses.
BIVLoss
(epsilon=1e-05)¶ - Description:
- Batch Inverse variance loss is defined as:
\[\mathcal{L}_{\mathrm{batch}}(D_i, \theta) = \left( \sum_{k=0}^K \dfrac{1}{\sigma_k^2 + \epsilon} \right)^{-1} \sum_{k=0}^K \dfrac{\mathcal{L}\left(f(x_k, \theta),\overline{y}_k\right)}{\sigma_k^2 + \epsilon} \label{eq:IVweighted_loss}\]-
forward
(y_pred, y, lbl_var)¶ - Description:
- Compute the forward pass for BIV loss function.