MetaShift

New framework tackles spurious correlations in deep learning models · 2 sources tracked

Researchers have developed a novel two-stage framework to improve the robustness of deep neural networks against distribution shifts by addressing spurious correlations. The method first uses generative intervention wit…