𝝀 \boldsymbol{\lambda} -Orthogonality Regularization for Compatible Representation Learning
Simone Ricci, Niccolò Biondi, Federico Pernici, Ioannis Patras, Alberto Del Bimbo · 2025
Read 𝝀 \boldsymbol{\lambda} -Orthogonality Regularization for Compatible Representation Learning by Simone Ricci, Niccolò Biondi, Federico Pernici, Ioannis Patras, Alberto Del Bimbo (2025) 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-2509-16664.txt.
Description
Simone Ricci and co-authors keep old and new image models in sync in a 2025 paper. Search systems turn images into lists of numbers for matching. When a model is retrained, its lists often clash with lists from the older model. Retraining everything is costly. The team add a training rule that keeps the lists compatible.
Their rule pushes certain directions in the lists toward right angles. Tests span many model types and datasets. The rule kept zero-shot skill intact, with no prior examples needed. It also let old and new lists work together after updates. Code is shared in public. Compatible lists could save firms from costly full rebuilds.
Catalog record
| CREATOR | Simone Ricci, Niccolò Biondi, Federico Pernici, Ioannis Patras, Alberto Del Bimbo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-16664