Recursive Inference Scaling: A Winning Path to Scalable Inference in Language and Multimodal Systems
Ibrahim Alabdulmohsin, Xiaohua Zhai · 2025
Read Recursive Inference Scaling: A Winning Path to Scalable Inference in Language and Multimodal Systems by Ibrahim Alabdulmohsin, Xiaohua Zhai (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-2502-07503.txt.
Description
Ibrahim Alabdulmohsin and Xiaohua Zhai propose running a model's layers in a recursive loop to get more from the same trained network. Their 2025 paper draws on findings that language has a fractal texture. Their method, RINS, adds recursive depth at use time rather than during training, and they compare it against more than fifty rival variants under matched computing budgets.
RINS beat the field, including a recent repeat-everything strategy. For a fixed model size and training cost, language scores rose clearly. The trick also carried to image-text systems, lifting zero-shot ImageNet accuracy for a known vision model by about two points. Scaling laws derived in the paper show the gains persist as data grows.
Catalog record
| CREATOR | Ibrahim Alabdulmohsin, Xiaohua Zhai |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-07503