Spectral Tempering for Embedding Compression in Dense Passage Retrieval Conference: Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval; July 20–24, 2026; 306 CCS: Information systems Retrieval models and ranking CCS: Computing methodologies Natural language processing
Yongkang Li, Panagiotis Eustratiadis, Evangelos Kanoulas · 2026
Read Spectral Tempering for Embedding Compression in Dense Passage Retrieval Conference: Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval; July 20–24, 2026; 306 CCS: Information systems Retrieval models and ranking CCS: Computing methodologies Natural language processing by Yongkang Li, Panagiotis Eustratiadis, Evangelos Kanoulas (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-2603-19339.txt.
Description
Li and co-authors shrink the vectors used in dense passage search without losing much quality. Cutting dimensions saves space, but classic methods waste capacity or boost noise. Their SpecTemp sets a scaling power for each target size from the data spectrum itself. Local signal-to-noise checks and a knee rule fix the value, with no labels needed. Across tests, scores stay near the best that a grid search can find. The method is learning-free and works across models. Code for the approach is public on GitHub.
Catalog record
| CREATOR | Yongkang Li, Panagiotis Eustratiadis, Evangelos Kanoulas |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24722 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.19339