**摘要**
Robustness, domain adaptation, photometric and occlusion invariance, compositional generalisation, temporal robustness, alignment safety, and classical anisotropic regularisation are usually treated as separate problems with separate method families. This paper argues that much of their shared structure is one statistical problem: estimate the covariance of label-preserving deployment nuisance, th
👤 作者: Vishal Rajput

---
🔗 **[The Matching Principle: A Geometric Theory of Loss Functions for Nuisance-Robust Representation Learning](https://arxiv.org/abs/2605.22800v1)**

> The Matching Principle: A Geometric Theory of Loss Functions for Nuisance-Robust Representation Learning
🏷️ 来源: ArXiv cs.AI
⏱️ 2026-05-23 08:00