Cost-Sensitive Self-Training for Optimizing Non-Decomposable Metrics Harsh Rangwani

However, the majority of work on self-training has focused on the objective of improving accuracy whereas practical machine learning systems can have complex goals (e.g.