mlr3torchAUM on CRAN

This project aims to extend the mlr3torch ecosystem by implementing AUC-oriented loss functions and training-time sampling strategies for imbalanced classification. The project will implement efficient Pairwise AUC loss functions (including multiple surrogate variants) and integrate them into mlr3torch via the existing t_loss() interface. In addition, over- and under-sampling strategies will be developed to improve class balance during training. The final outcome will be a well-tested, CRAN-compliant implementation with documentation, enabling users to easily apply AUC-based optimization and improved sampling methods within mlr3 workflows.