The Sketched Isotropic Gaussian Regularizer (SIGReg) prevents representation collapse by ensuring all learned semantic dimensions remain statistically independent and aligned with the predictive objective

The Sketched Isotropic Gaussian Regularizer (SIGReg) ensures that all learned semantic dimensions remain statistically independent and aligned with the predictive objective, preventing collapse even as the architecture grows

http://arxiv.org/abs/2608.27367v1