The Matching Principle: A Geometric Theory of Loss Functions for Nuisance-Robust Representation Learning
Vishal Rajput · 2026
Read The Matching Principle: A Geometric Theory of Loss Functions for Nuisance-Robust Representation Learning by Vishal Rajput (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-22800.txt.
Description
Rajput states a Matching Principle for training penalties. First name the deployment directions a model must resist. Then ask whether the training penalty covers them. In a linear ridge model, the default even-spread penalty cuts deployment drift with no coverage miss by construction. A penalty aimed at the wrong axes leaves a residual floor. Across seven domains, one named second-moment penalty beat unregularized training. The lesson: control representation sensitivity by default.
Catalog record
| CREATOR | Vishal Rajput |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 150679 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.22800