Loss Balancing

为复现：　Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics

(https://arxiv.org/pdf/1705.07115.pdf)

文章中：

原文章定义sigma为噪声，在原论文复现中(https://github.com/yaringal/multi-task-learning-example/blob/master/multi-task-learning-example.ipynb)直接是可学习的参数,见下图：

用pytorch复现的坑：

１、要用parameter,而不是定义常数torch.Tensor(torch.zeros(1))

from torch.nn import Parameter
    self.log_cate    =  Parameter(torch.Tensor(1))
    self.log_regress = Parameter(t