Quantifying Distribution Shifts and Uncertainties for Enhanced Model Robustness in Machine Learning Applications

arXiv:2405.01978v1 Announce Type: new
Abstract: Distribution shifts, where statistical properties differ between training and test datasets, present a significant challenge in real-world machine learning applications where they directly impact model generalization and robustness. In this study, we explore model adaptation and generalization by utilizing synthetic data to systematically address distributional disparities. Our investigation aims to identify the prerequisites for successful model adaptation across diverse data distributions, while quantifying the associated uncertainties. Specifically, we generate synthetic data using the Van der …

abstract applications arxiv challenge cs.lg data datasets distribution explore impact machine machine learning machine learning applications model adaptation model generalization model robustness robustness statistical stat.ml study synthetic synthetic data test test datasets training type world