This is my blog.

# Loss weight

## Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics——2018

Make the observation that the performance of such systems is strongly dependent on the relative weighting between each task’s loss. 这篇文章主要讲述在多任务回归和分类中，多任务联合学习可以提升各任务的学习效果，因为多任务可以共享数据集、共享低层特征、减少计算，但是loss中如何分配相对权重也是重要的问题之一。

Uncertainty:

• Epistemic uncertainty偶然不确定性: lack of training data
• Aleatoric uncertainty认知不确定性: all explanatory variables with increasing precision.
• Data-dependent or Heteroscedastic uncertainty: depends on the input data输入数据没有在训练数据中见过

Softmax:

$\sigma_2$趋向于1时，三式取等号。