Stability Evaluation via Distributional Perturbation Analysis

arXiv:2405.03198v1 Announce Type: cross
Abstract: The performance of learning models often deteriorates when deployed in out-of-sample environments. To ensure reliable deployment, we propose a stability evaluation criterion based on distributional perturbations. Conceptually, our stability evaluation criterion is defined as the minimal perturbation required on our observed dataset to induce a prescribed deterioration in risk evaluation. In this paper, we utilize the optimal transport (OT) discrepancy with moment constraints on the \textit{(sample, density)} space to quantify this perturbation. Therefore, our stability …

abstract analysis arxiv criterion cs.lg dataset deployment environments evaluation math.oc performance risk sample stability stat.ml type via