LoUQAL: Low-fidelity informed Uncertainty Quantification for Active Learning in the chemical configuration space
Vivin Vinod, Peter Zaspel · 2025
Read LoUQAL: Low-fidelity informed Uncertainty Quantification for Active Learning in the chemical configuration space by Vivin Vinod, Peter Zaspel (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-2508-15577.txt.
Description
Training a machine learning model for chemistry is throttled by the cost of accurate quantum calculations. Active learning should pick the most useful molecules to calculate, yet in practice it often does worse than random picking. Vivin Vinod and Peter Zaspel diagnose the flaw in a 2025 paper: standard active learning chases the model's uncertainty, its variance, while error also comes from bias.
Their method, LFaB, estimates bias cheaply by using low-fidelity calculations, rough answers that cost little, as a stand-in for the truth. Picking samples to cut bias nearly matched the best possible greedy selection in their tests. On quantum chemistry tasks such as excitation energies and energy surfaces, the method cut training data needs by up to tenfold against standard active learning. The paper reframes what a smart sampler should hunt for.
Catalog record
| CREATOR | Vivin Vinod, Peter Zaspel |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-15577