Scaling laws in machine learning on non-ideal experimental data enable inverse design of radiative-cooling coatings
Source check: Source checked. Not independent scientific confirmation.
Finding: Machine learning on 11,520 experimentally fabricated radiative-cooling coatings showed power-law decreases in prediction error as training data increased. Experimentally tested inverse designs matched a broadband-emitter target more closely with larger datasets, but did not outperform the best original sample.
Technology: Forward and inverse neural networks; spectrum-conditioned diffusion model
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
- AI contribution
- Learned forward/inverse models and a conditional diffusion model mapped measured coating spectra and parameters for prediction and inverse design.
- Test and result
- The automated platform produced and optically characterized 11,520 coatings across nine material/process parameters. Models showed power-law error reduction with more training data; selected inverse designs were experimentally tested for matching a broadband-emitter target.
- Limitation / still unknown
- The selected designs did not outperform the best original sample; claims are limited to explored compositions, processing and measurement setup.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2611.19596v1 · https://export.arxiv.org/api/query?id_list=2611.19596&max_results=1 · https://arxiv.org/pdf/2611.19596v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 22 Nov 2026. Review scope: Primary result-bearing material for the stated method, result and evaluation, with bibliographic version information..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.