One-Step Generalization Ratio Guided Optimization for Domain Generalization
Researchers have introduced GENIE, a novel optimizer designed to improve domain generalization in machine learning models. GENIE utilizes the One-Step Generalization Ratio (OSGR) to dynamically adjust parameter updates, preventing over-reliance on domain-specific features and promoting the learning of domain-invariant characteristics. This approach aims to balance parameter contributions and gradient alignment, theoretically maintaining SGD's convergence rate while empirically outperforming existing optimizers. AI
IMPACT Introduces a novel optimization technique to improve model generalization across different datasets.