Theuseofmetric elicitation enables a practitioner to tune the performance and fairness metrics to the task, context,andpopulationathand.

Constructing a directed cyclic graph (DCG) is challenged by both algorithmic difficulty and computational burden. Comparing multiple DCGs is even more difficult, compounded by the need to identify dynamic causalities across graphs.

Deep neural networks (DNNs) are the de facto methodology in many artificial intelligence areas nowadays [25, 37]. How to effectively train a deep network has been a central topic for decades.

Why do some DA schemes ( e.g., CutMix) inherently perform much better than

We also establish a global superlinear convergence rate ofO((1t)t).