multi-task learning (MTL) has achieved great success in various research
domains, such as CV, NLP and IR etc. Due to the complex and competing task
correlation, na\"ive training all tasks may lead to inequitable learning,
\textit{i.e.} some tasks are learned well while others are overl