Can Less Precise Be More Reliable? A Systematic Evaluation of Quantization’s Impact on CLIP Beyond Accuracy
Aymen Bouguerra, Daniel Montoya, Alexandra Gomez-Villa, Fabio Arnez Chokri Mraidha · 2025
Read Can Less Precise Be More Reliable? A Systematic Evaluation of Quantization’s Impact on CLIP Beyond Accuracy by Aymen Bouguerra, Daniel Montoya, Alexandra Gomez-Villa, Fabio Arnez Chokri Mraidha (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-2509-21173.txt.
Description
Aymen Bouguerra and 4 co-authors wrote a machine learning paper in 2025 on vision and language models. They study CLIP, a model that links images and text. Such models cost a lot to run, so users often shrink them by quantization. That step stores weights with less precision. The paper asks if less precision hurts trust in the model.
The team ran a very large set of tests across many settings. They found that shrinking can raise accuracy, improve how well scores match real rates, and help spot odd inputs. It did not help with shifted inputs or false links. They trace the gain to the model leaning more on strong, simple features and less on weak noise.
Catalog record
| CREATOR | Aymen Bouguerra, Daniel Montoya, Alexandra Gomez-Villa, Fabio Arnez Chokri Mraidha |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 86 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-21173