Nested Hash Layer: A Plug-and-play Module for Multiple-length Hash Code Learning
Liyang He, Yuren Zhang, Rui Li, Zhenya Huang, Runze Wu, Enhong Chen · 2024
Read Nested Hash Layer: A Plug-and-play Module for Multiple-length Hash Code Learning by Liyang He, Yuren Zhang, Rui Li, Zhenya Huang, Runze Wu, Enhong Chen (2024) 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-2412-08922.txt.
Description
Liyang He and co-authors sped up image search systems that compress pictures into short binary hash codes. In deep supervised hashing, a model learns codes whose similarity mirrors image similarity, making huge collections searchable. Short codes are fast but less accurate, long codes the reverse, and finding the best length meant training separate models. Their Nested Hash Layer plugs into existing models and produces several code lengths at once, arranged like nesting dolls.
Two additions make the nesting work. A weighting scheme balances the competing training goals of each length, and a self-distillation step lets long codes teach the shorter ones. Across several base models the module cut total training time roughly five to eight fold while raising average accuracy by about three percent.
Catalog record
| CREATOR | Liyang He, Yuren Zhang, Rui Li, Zhenya Huang, Runze Wu, Enhong Chen |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-08922