FAAST: Forward-Only Associative Learning via Closed-Form Fast Weights for Test-Time Supervised Adaptation
Guangsheng Bao, Hongbo Zhang, Han Cui, Ke Sun, Yanbin Zhao, Juncai He, Yue Zhang · 2026
Read FAAST: Forward-Only Associative Learning via Closed-Form Fast Weights for Test-Time Supervised Adaptation by Guangsheng Bao, Hongbo Zhang, Han Cui, Ke Sun, Yanbin Zhao, Juncai He, Yue Zhang (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-2605-04651.txt.
Description
Guangsheng Bao and six co-authors propose a forward-only method for adapting a pretrained model to a new supervised task at deployment time. Standard adaptation either pays the high cost of backpropagation training or carries heavy memory and context overheads from example-based methods.
Their method compiles labelled examples directly into fast weights. It uses a closed-form calculation in a single forward pass. There is no dependence on memory or context afterwards. Inference therefore runs in constant time. Across image classification and language modelling benchmarks, it matched or beat backpropagation-based adaptation. It cut adaptation time by more than ninety percent. It also used far less memory than context-based rivals.
Catalog record
| CREATOR | Guangsheng Bao, Hongbo Zhang, Han Cui, Ke Sun, Yanbin Zhao, Juncai He, Yue Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.04651