Physics-guided active learning and XGBoost screening are combined in the design of functional thermoelectric composites

AI editorPublished 2026.10.06Posted 2026.10.07 18:02 KST

Physics-guided active learning combining Bayesian optimization, XGBoost regression, and RFE was used to predict interface barriers from literature-based data. The approach narrowed candidates to 5 composite systems and led to experimental validation, with a test R² of 0.87 and, in PEDOT:PSS/EGaIn, a 0.24 eV barrier and an 8.80 mV open-circuit voltage at 25 K being reported. [1]

1/ AI utilization method: physics-guided active learning combining Bayesian optimization, XGBoost regression, and RFE
2/ Applied task: predict interface barriers from literature-based data, select candidate composite systems, and then proceed to synthesis and experimental validation
3/ Reported results: test R² 0.87, and a 0.24 eV barrier and 8.80 mV open-circuit voltage in PEDOT:PSS/EGaIn
4/ This study used physics-guided active learning for the energy-filtering design of flexible thermoelectric composites.
5/ In the process of predicting interface barriers based on literature-derived data, Bayesian optimization, XGBoost regression, and recursive feature elimination (RFE) were combined to narrow the candidates.
6/ Reported results included an R² of 0.87 on the test set, and among the 5 selected candidate composite systems, the PEDOT:PSS/EGaIn system showed a 0.24 eV barrier close to the predicted value and an open-circuit voltage of 8.80 mV at a temperature gradient of 25 K.
7/ This can be viewed as an active-learning-style materials design case linking prediction, selection, synthesis, and experimental validation.
8/ The significance for materials research is that literature data and physical intuition can be used together to reduce the candidate space and prioritize combinations with high experimental cost.
9/ However, generalization to other material systems requires separate confirmation because of literature-data bias, the small number of validated candidates, and tuning specific to particular interface combinations. [1]

How to apply this

You can predict key interfacial properties of candidate materials from literature data and prioritize experimental combinations through active learning.

Organize interfacial-barrier-related variables from literature and internal experiments into a training table.
Use XGBoost and feature selection to reduce important explanatory variables, then calculate an active-learning ranking.
Synthesize only a small number of top candidates and measure their interfacial barriers and electrical responses.
Compare predicted and measured values and update the exploration candidates for the next round.

Data needed: A dataset of composition, processing, and interfacial-property information extracted from literature, along with a small number of synthesized and measured samples for validation, is needed.
Checking the result: Synthesize the selected candidates and compare predicted interfacial properties with measured values, then verify regression performance and uncertainty on an independent test set.
What can go wrong: Because of missing values and bias in literature data, and the small validation scale such as 5 candidates, it is difficult to assume that the reported performance will translate directly into real search efficiency.

이 가이드는 원문의 방법을 바탕으로 AI가 제안한 적용 방향입니다. 원문에 적힌 내용이 아니며 같은 결과가 재현된다는 보장도 아닙니다.

웹 자료를 바탕으로 AI가 자동 요약한 글입니다. 연구의 세부 조건과 결과는 연결된 원문을 확인해주세요.

Physics-guided active learning and XGBoost screening are combined in the design of functional thermoelectric composites

How to apply this

Related concepts and practice

Sources