From Objectives to What Models Learn: A Landau Theory of Invariant Learning

Pinli Wang, Yue He, Peng Cui

Invariant learning seeks representations that remain predictive across environments, yet the behavior of its objectives along the regularization path is often opaque. We address th…