Efficient Multi-class Classification with Neyman-Pearson Error Control

The authors propose two algorithms, NPMC-CX and NPMC-ER, to solve the Neyman-Pearson multi-class classification problem, which aims to minimize a weighted sum of misclassification errors while controlling the error rates for specific classes. The algorithms leverage the connection between the Neyman-Pearson problem and cost-sensitive learning, and provide theoretical guarantees on the multi-class NP oracle properties.