GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networks

GradNorm: 使用梯度标准化调整多任务损失权重.

Loss-Balanced Task Weighting to Reduce Negative Transfer in Multi-Task Learning

通过损失平衡任务加权解决多任务学习中的负迁移.

End-to-End Multi-Task Learning with Attention

多任务注意力网络与动态权重平均.

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

使用同方差不确定性调整多任务损失权重.

Searching for Activation Functions

Swish：自动搜索得到的一种自门控的激活函数.

The Quest for the Golden Activation Function

ELiSH：使用遗传算法寻找最优激活函数.