Researchers have developed a novel reinforcement learning framework to accelerate the discovery of perovskite solar cell recipes. This method uses a surrogate model, Tabular Prior-data Fitted Network (TabPFN), to guide the generation of new candidate compositions and device layers. Evaluations on a dataset of 1,846 devices demonstrated that this approach can significantly increase the mean power conversion efficiency of selected top-ten recipes from 19.68% to 23.74%. The study compared several reinforcement learning policies, finding that Proximal Policy Optimization (PPO) and its recurrent variant (RPPO) outperformed baseline methods in generating promising candidates within a limited evaluation budget. AI
IMPACT This research demonstrates a novel application of reinforcement learning for accelerating materials science discovery, potentially speeding up the development of new solar cell technologies.
RANK_REASON The cluster contains an academic paper detailing a new method for scientific discovery using AI. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- AX Boötis
- L-SHADE
- perovskite solar cell
- Proximal Policy Optimization
- recurrent PPO
- reinforcement learning
- RpLP0
- TabPFN
- Tabular Prior-data Fitted Network
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →