Modular Multi-Task Learning for Chemical Reaction Prediction
JiayunPang, Ahmed M.Zaitoun, Xacobe CousoCambeiro, IvanVulić · 2026
Read Modular Multi-Task Learning for Chemical Reaction Prediction by JiayunPang, Ahmed M.Zaitoun, Xacobe CousoCambeiro, IvanVulić (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-2602-10404.txt.
Description
Adapting broad organic-chemistry language models to smaller, domain-specific reaction data sets is a key challenge: new reaction knowledge must be learned without losing general chemical understanding. A 2026 paper led by Jiayun Pang, with three co-authors, tests low-rank adaptation, a light-touch alternative to full fine-tuning, for reaction prediction on limited, complex data sets. Using United States patent reaction classes and hard carbon-hydrogen functionalization reactions, they test forward prediction, retrosynthesis, and reagent prediction.
Low-rank adaptation matches full fine-tuning on accuracy and better preserves multi-task work, curbing catastrophic forgetting. Both approaches work beyond their training data, producing plausible alternative solvent picks. Fine-tuning on carbon-hydrogen functionalization shows the two approaches encode different reactivity patterns, suggesting low-rank methods may adapt better to specific reactions. The results support modular, light-touch fine-tuning as a practical route for chemistry work.
Catalog record
| CREATOR | JiayunPang, Ahmed M.Zaitoun, Xacobe CousoCambeiro, IvanVulić |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-10404