arXiv：cs.LG· Qiwei Han, Chi Zhou·· 11 小时前
钙钛矿光伏机器学习评估的溯源分组协议
Provenance-grouped protocol for evaluating machine learning in perovskite photovoltaics
AI 导读
针对钙钛矿太阳能电池机器学习模型评估中随机数据划分导致的信息泄漏问题，研究者基于 Perovskite Database 的机构溯源信息，建立了逐级隔离重复器件、论文和机构的分组划分层级。
正文
Abstract:Machine-learning models for perovskite solar cells are typically evaluated using random data splits, which we show systematically leak information through massive internal duplication. Because nearly two-thirds of curated devices share their exact recipe with another, random splits primarily reward the mere recall of historical data rather than testing true physical generalisation. To solve this, we mapped the institutional provenance of the Perovskite Database and established a rigorous hierarchy of grouped splits that progressively isolate duplicate devices, publications, and institutions. Benchmarking thirteen models establishes new state-of-the-art performance across all splits and reveals a critical regime shift: while tree ensembles dominate random splits by recognising familiar recipes, their ranking collapses on genuinely unseen institutions and future discoveries, where composition-aware neural architectures excel. Furthermore, feature importance demonstrates that the device architecture and fabrication methodology dictate final efficiency just as strongly as the absorber chemistry. This standardised protocol empowers the community to separate memorisation from true physical discovery.
| Comments: | perovskite solar cells, provenance-grouped protocol, state of art, benchmark |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.12766 [cs.LG] |
| (or arXiv:2610.12766v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.12766 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Qiwei Han [view email]
[v1]
Thu, 8 Oct 2026 22:17:48 UTC (1,261 KB)
来源：arXiv：cs.LG · arxiv.org